Internship Report

Chapters Table of Contents

1. About the Company
 1.1 Company Profile
 1.2 Industry Focus

2. About the Domain

 2.1 Artificial Intelligence
 2.2 Data Science
 2.3 Applications
 2.4 Future Scope

3. Tools and Technologies

 3.1 Python
 3.2 Jupyter Notebook
 3.3 Libraries (Pandas, NumPy, Scikit-learn, etc.)

4. Tasks Performed
 4.1 Week-wise Summary
 4.2 Glimpse of Project Execution

5. Mini Projects
 5.1 Handwritten Digits Recognition
 5.2 FIFA 20 Player Analysis
 5.3 PUBG Match Winner Prediction
 5.4 Heart Disease Prediction
 5.5 Outputs
6. Conclusion

Dept. of AI&ML, GNDECB Page |4

Internship Report

CHAPTER 1:

ABOUT THE COMPANY

1.1 COMPANY PROFILE

Aspect Details

Organization
Rubixe™
Name

Organization
Logo

Established 2015

Type Private Technology Company

Ownership Privately Held

Headquarters Bangalore, Karnataka, India

- Artificial Intelligence Solutions

- Business Automation
Key Functions
- Data Science & Analytics Consulting
- AI-based Product Development

- Predictive Analytics
Product
- AI-Driven Business Intelligence
Specialization
- Custom ML and NLP Tools

- Retail
Key Domains - Finance
Served - Healthcare
- Manufacturing

Dept. of AI&ML, GNDECB Page |5

Internship Report

Aspect Details

- AI Chatbots for customer support

Notable Projects - Fraud detection systems
- Personalized recommendation engines

- AI and automation consulting for clients across India, the Middle

Global Role
East, and parts of Europe

To deliver intelligent and adaptive technology that transforms

Vision
businesses through innovation.

To enable businesses to leverage the power of disruptive

Mission technologies like AI, Machine Learning, and IoT to drive digital
transformation.

Fig 1.1.1 Company Building

Dept. of AI&ML, GNDECB Page |6

Internship Report

Fig 1.1.2 Company Lobby

Fig 1.1.3 Company Break Room

Dept. of AI&ML, GNDECB Page |7

Internship Report

Fig 1.1.4 Company Work Space

Fig 1.1.5 Company Work Space

Dept. of AI&ML, GNDECB Page |8

Internship Report

1.2 Industry Focus

Rubixe serves clients across a broad spectrum of industries, reflecting the versatility of its
AI-driven solutions. The company’s industry focus is not limited to one sector; instead, it
includes major domains such as Healthcare, Finance and Banking, Information
Technology, Retail, Manufacturing, Education, Agriculture, and [Link]
[Link]. For instance, in healthcare, Rubixe might develop predictive analytics for
patient data, whereas in finance it could implement AI for fraud detection or algorithmic
trading support. In retail and e-commerce, Rubixe’s solutions could involve personalized
recommendation systems and inventory optimization, while in manufacturing, they might
focus on predictive maintenance and automation of assembly lines. This wide industry
coverage (from startups to large enterprises) indicates Rubixe’s adaptive approach in
applying AI to different contexts. The company’s team includes domain experts who
understand the unique challenges of each sector, ensuring that AI solutions are effective
and relevant to the client’s field. By maintaining a diverse industry focus, Rubixe not only
expands its market reach but also enriches its expertise, as lessons learned in one domain
can often inspire innovative applications in another.

Dept. of AI&ML, GNDECB Page |9

Internship Report

CHAPTER 2:

ABOUT THE DOMAIN

This chapter provides an overview of the technical domain central to the internship:
Artificial Intelligence and Data Science. It covers the fundamental concepts of AI and
data science, discusses some of their real-world applications, and reflects on the future
scope of these fields. Understanding the domain is crucial, as the internship work involved
applying AI and data science techniques to solve practical problems in various projects.
The following sections define Artificial Intelligence and Data Science, outline key
application areas, and consider emerging trends and future developments.

2.1 Artificial Intelligence

Artificial Intelligence (AI) refers to the simulation of human intelligence processes by

machines, especially computer [Link]. In essence, AI enables computers
to perform tasks that would typically require human intelligence, such as learning from
experience, understanding language, recognizing patterns, and making decisions. Core
subfields of AI include machine learning (where algorithms improve through data
exposure), natural language processing (enabling machines to understand and generate
human language), computer vision (interpreting visual information from the world), and
expert systems (which use logic and rules to mimic human experts)[Link]. AI
systems can be either narrow AI (designed for a specific task, like a speech recognition
assistant or a chess-playing program) or general AI (an as-yet unachieved state where a
machine possesses broad intelligence comparable to a human). Over the years, AI
techniques have advanced significantly, enabling the development of systems that can
learn, perceive, and adapt to new information. During the internship projects, AI concepts
were applied, for example, in training models to recognize handwriting and predict
outcomes, demonstrating the practical use of algorithms that enable machine intelligence.

Dept. of AI&ML, GNDECB P a g e | 10

Internship Report

2.2 Data Science

Data Science is an interdisciplinary field focused on extracting knowledge and insights

from large data sets and applying those insights to solve complex problems
[Link]. It blends techniques from statistics, computer science, and domain
expertise to collect, process, and analyze data. Key components of data science include
data collection (gathering relevant data from various sources), data wrangling or
preprocessing (cleaning and transforming raw data into a usable format), exploratory data
analysis (using statistical methods and visualization to discover patterns or anomalies in
the data), and modeling (building predictive or descriptive models using machine learning
or statistical algorithms). Data science also involves validating models and interpreting
results to inform decision-making. A data scientist must not only build accurate models but
also understand the context of the data—this is where domain knowledge comes into play.
In this internship, data science practices were central: each project began with
understanding and exploring the dataset at hand (whether images of digits, football player
statistics, game outcomes, or medical records) and then applying appropriate modeling
techniques to draw conclusions or make predictions. The interdisciplinary nature of data
science was evident, as success required a mix of programming skills (for implementation
in Python), mathematical understanding (for algorithm correctness), and contextual
knowledge (to make sense of results in domains like sports or healthcare).

2.3 Applications

AI and data science have a wide array of applications across different industries, many of
which are becoming integral to how modern organizations operate. A few notable
application areas include:

 Healthcare: AI and data science are used for diagnosing diseases, predicting
patient outcomes, and personalizing treatment plans. For example, machine
learning models can analyze medical images (X-rays, MRIs) to detect anomalies or
predict the likelihood of conditions such as cancer or heart disease. In this
internship’s Heart Disease Prediction project, the application of data science in

Dept. of AI&ML, GNDECB P a g e | 11

Internship Report

healthcare was evident, as a model was developed to predict cardiac risk based on
patient data. Such models can assist doctors in early detection of high-risk patients,
potentially saving lives by enabling proactive care.
 Finance: In the financial industry, AI algorithms detect fraudulent transactions,
assess credit risk, automate trading strategies, and provide customer service through
chatbots. Data science enables banks and financial firms to analyze vast quantities
of transactions and client data to find patterns indicative of fraud or to segment
customers for tailored services. The ability of AI to learn from historical financial
data helps in making predictions about market trends or default probabilities,
thereby supporting more informed and efficient financial decision-making.
 Retail and E-commerce: Businesses leverage AI for recommendation systems
(suggesting products to customers based on their browsing and purchase history),
inventory management, and demand forecasting. Data science techniques analyze
customer behavior data to optimize pricing strategies and marketing campaigns.
For instance, an AI system can predict which products are likely to be popular in
the next season and ensure they are stocked appropriately.
 Transportation and Automotive: A high-profile application of AI is in
autonomous vehicles (self-driving cars) where computer vision and deep learning
are used to navigate roads and make real-time decisions. Additionally, AI optimizes
logistics and supply chain operations by predicting the fastest delivery routes and
minimizing fuel consumption.
 Manufacturing: AI-driven predictive maintenance systems analyze sensor data
from machinery to predict failures before they occur, thus avoiding costly
downtime. Robotics powered by AI also enable automation of complex
manufacturing tasks with precision and adaptability.
 Entertainment and Sports Analytics: Streaming services and media companies
use data science to analyze viewer preferences and recommend content. In sports,
AI and data analysis are used to evaluate player performance and devise game
strategies. One of the internship projects, the FIFA 20 Player Analysis, mirrors real-
world sports analytics where player data is examined to glean insights on
performance and value.

Dept. of AI&ML, GNDECB P a g e | 12

Internship Report

These examples barely scratch the surface; AI and data science are also transforming
education (through personalized learning platforms), agriculture (through crop and soil
monitoring), and government (through smart cities and policy planning), among others. In
every application, the goal is to harness data to make better decisions, automate tasks, and
create intelligent behavior in software or machines. The projects completed during the
internship each correspond to a real-world application of AI: computer vision in the digit
recognition task, sports analytics in the FIFA data task, game outcome prediction in the
PUBG task, and medical risk prediction in the heart disease task. This demonstrates the
versatility of AI and data science techniques across domains.

2.4 Future Scope

The future of Artificial Intelligence and Data Science is both exciting and expansive. As
computational power grows and algorithms become more sophisticated, AI systems are
expected to achieve even higher levels of performance and autonomy. One emerging trend
is the development of more generalized AI models (such as large language models and
advanced neural networks) that can perform a wider range of tasks and learn with less
human supervision. These models are increasingly capable of understanding context and
generating human-like responses, which could revolutionize fields like customer service,
content creation, and beyond.

In the realm of data science, the volume of data generated globally is increasing
exponentially (often referred to as the data deluge). This creates a growing opportunity to
extract insights, but also necessitates advancements in big data technologies, cloud
computing, and data engineering to handle and process data efficiently. The future will
likely see data science integrated even more into business decision processes, with real-
time analytics and dashboards guiding strategic and operational choices in organizations.

Another important aspect of the future of AI is its integration with IoT (Internet of Things)
and edge computing. As more devices become smart and interconnected, AI algorithms
will be deployed on distributed devices (from smartphones to industrial sensors) to provide

Dept. of AI&ML, GNDECB P a g e | 13

Internship Report

instant insights and automation at the source of data. This could enable smarter homes,
smarter cities, and responsive industrial systems.

In terms of societal impact, AI and data science are poised to significantly influence the
job market and skill requirements. There is an increasing demand for professionals skilled
in AI/ML, and educational curricula are evolving to include these topics from early stages.
However, the rise of AI also brings challenges that will shape its future: ethical
considerations and governance. Issues such as data privacy, algorithmic bias, and
transparency of AI decisions are under scrutiny. Future AI systems will need to be
developed with fairness and accountability in mind, leading to the growth of fields like AI
ethics and explainable AI.

Overall, the scope of AI and data science is set to expand into virtually every field.
Continuous research and development are making AI solutions more powerful and
accessible. For a practitioner like an intern entering this field, it means the learning is
ongoing—new tools, techniques, and best practices emerge regularly. The projects done
during the internship provide a foundation, but staying updated will be crucial. In the near
future, one can expect more automation of routine tasks, more accurate predictive models
improving decision-making, and AI tackling complex problems (like climate modeling or
drug discovery) that were previously beyond reach. The trajectory suggests that AI and
data science will remain at the forefront of technological innovation, driving progress in
the coming years.

Dept. of AI&ML, GNDECB P a g e | 14

Internship Report

CHAPTER 3:

TOOLS AND TECHNOLOGIES USED

A variety of software tools, programming platforms, and libraries were used throughout
the internship to accomplish the tasks and projects. As the internship work centered on data
analysis and machine learning, the tools chosen are standard in the data science community.
This chapter describes the main tools and technologies employed, namely the Python
programming language, the Jupyter Notebook environment, and several important Python
libraries such as Pandas, NumPy, Scikit-learn, etc. Each of these played a critical role in
enabling the intern to implement solutions efficiently and effectively.

3.1 Python

Fig 3.1.1 Python interface

Python was the primary programming language used during the internship. Python is a
high-level, interpreted language known for its simplicity and readability, which makes it
particularly well-suited for data science and AI projects. One of the key advantages of
Python is its extensive collection of libraries and frameworks for scientific computing and
machine learning (for example, Pandas for data manipulation, or TensorFlow/PyTorch for
deep learning). Python’s syntax is concise and clear, allowing developers to prototype ideas
quickly with relatively few lines of code. During the internship projects, Python was used

Dept. of AI&ML, GNDECB P a g e | 15

Internship Report

for tasks such as loading and preprocessing datasets, implementing machine learning
algorithms, and evaluating model performance. The language’s interactive nature
(especially when used in notebooks) helped in iteratively developing solutions: the intern
could write a portion of code, run it, inspect the output, and adjust as needed in a rapid
development cycle. Additionally, Python’s strong community support and documentation
meant that any issues encountered could be resolved by referencing existing resources.
Overall, Python’s role was foundational, providing the programming backbone for all
analyses and experiments conducted in the internship.

3.2 Jupyter Notebook

Fig 3.2.1 Jupyter interface

The Jupyter Notebook was the primary development environment for the projects. Jupyter
Notebook is an open-source web-based interactive platform that allows developers to
combine live code, visualizations, and narrative text in a single document. This format was
extremely useful for the internship, as it facilitated a literate programming approach: one
can execute code step by step and document the thought process alongside the results. Each
project was organized in a dedicated Jupyter notebook, as per the internship requirements.
Using notebooks, the intern performed data exploration by writing code to, for example,

Dept. of AI&ML, GNDECB P a g e | 16

Internship Report

display the first few rows of a dataset or plot distributions, and then immediately viewed
the results (tables, charts) inline. The notebooks also made it convenient to tune machine
learning models and record observations; for instance, the intern could run multiple training
experiments in sequence and compare outcomes all within one notebook environment.
Another advantage is the ease of presentation—completed notebooks with outputs and
markdown explanations can serve as final reports or records of the work. This was helpful
for sharing progress with mentors and for final submission, as the notebook contained both
the code and the interpretation of results (including any challenges faced and how they
were addressed). The interactive and incremental nature of Jupyter notebooks significantly
enhanced productivity and clarity during the internship.

3.3 Libraries (Pandas, NumPy, Scikit-learn, etc.)

Several specialized Python libraries were utilized to carry out data science tasks efficiently.
The major libraries and frameworks used during the internship include:

 Pandas: This library was used for data manipulation and analysis. Pandas provides
the DataFrame structure, which is excellent for handling tabular data (similar to
spreadsheets or SQL tables). With Pandas, the intern could easily load datasets
(from CSV files, for example), clean data (handle missing values, filter rows, create
new columns), and perform aggregations or summary statistics. Throughout the
projects, Pandas was essential for preparing data before feeding it into machine
learning models—such as selecting relevant features for the FIFA player data or
merging information in the PUBG match data.
 NumPy: NumPy is the fundamental package for scientific computing with Python,
providing support for large, multi-dimensional arrays and matrices, along with a
collection of mathematical functions to operate on these arrays. In the context of
this internship, NumPy was frequently used under the hood (many other libraries
like Pandas and Scikit-learn are built on NumPy arrays for performance). Directly,
it was used for tasks such as numerical computations, generating random numbers
(e.g., splitting data into random train/test sets before using library functions), and
implementing any algorithmic logic that required array manipulations. NumPy’s

Dept. of AI&ML, GNDECB P a g e | 17

Internship Report

efficiency in handling computations made operations on large datasets (like

thousands of game records or image pixel arrays) feasible and fast.
 Scikit-learn: This is a powerful machine learning library that was central to
building and evaluating models for the projects. Scikit-learn provides a consistent
interface for a wide range of machine learning algorithms—for classification,
regression, clustering, dimensionality reduction, etc. During the internship, Scikit-
learn was used to implement models such as Support Vector Machines (SVM), K-
Nearest Neighbors (KNN), decision trees, random forests, logistic regression, and
clustering algorithms like K-Means. It also offers utilities for splitting data, cross-
validation, and computing evaluation metrics. For example, in the Handwritten
Digits project, Scikit-learn’s SVM and KNN classifiers were used to recognize
digits, and accuracy scores were obtained easily through its API. Similarly, in the
Heart Disease project, a logistic regression model (from Scikit-learn) might have
been used to predict disease presence, along with metrics like precision and recall
to gauge performance. The library’s reliability and ease of use significantly
streamlined the experimentation with different algorithms.
 Matplotlib and Seaborn: These libraries were used for data visualization.
Matplotlib is a plotting library that allows the creation of static, interactive, and
animated visualizations in Python. Seaborn is built on Matplotlib and provides a
higher-level interface for drawing attractive statistical graphics. Visualizing data
was a critical part of the internship projects, especially during exploratory analysis
and result interpretation. For instance, using Matplotlib/Seaborn, the intern created
histograms of player ages, scatter plots of player rating versus age for the FIFA
analysis, and correlation heatmaps of features in the Heart Disease dataset. These
visuals helped in understanding data distributions, identifying trends, and
presenting findings.
 TensorFlow/Keras (and other frameworks, if applicable): In some cases, more
advanced frameworks like Keras (which uses TensorFlow as a backend) were
utilized for building neural network models. This was particularly relevant to the
Handwritten Digits Recognition project, where a simple neural network (multilayer
perceptron) was trained to improve accuracy on classifying digit images. Keras

Dept. of AI&ML, GNDECB P a g e | 18

Internship Report

provides a user-friendly way to define and train deep learning models. While the
internship primarily focused on classical machine learning via Scikit-learn, the use
of Keras for the digit classification task introduced the intern to the basics of deep
learning (such as constructing neural network layers and using an optimizer to
adjust weights based on training data).

In summary, these libraries formed the toolkit that enabled effective handling of all aspects
of the projects—from reading and exploring data (Pandas, NumPy) to building predictive
models (Scikit-learn, TensorFlow) and visualizing results (Matplotlib, Seaborn). Mastery
of these tools is fundamental for any data science project, and the internship provided
ample opportunity to apply and strengthen skills in using each of them.

Dept. of AI&ML, GNDECB P a g e | 19

Internship Report

CHAPTER 4:

TASKS PERFORMED

The internship was structured over 15 weeks, and each week had specific objectives and
deliverables. This section provides a week-wise summary of the tasks performed, the
learning activities, and the progress made throughout the internship. The work gradually
progressed from introductory orientation and training in the initial weeks to more complex
project work in the later weeks. By breaking down the experience week by week, it
becomes clear how the intern’s responsibilities evolved and how the mini projects were
distributed across the internship timeline. The summary below highlights key activities and
accomplishments for each week.

4.1 Week-wise Summary

Dept. of AI&ML, GNDECB P a g e | 20

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 1
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 17-02- 10:00 05:00 Worked on handwritten digits Consistently delivered
2025 AM PM recognition – exploring MNIST dataset tasks for week 1, with
and loading into notebook initiative and attention to
detail.

2 18-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – preprocessing images tasks for week 1, with
and visualizing data initiative and attention to
detail.

3 19-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – applying basic classifiers tasks for week 1, with
like KNN and SVM initiative and attention to
detail.

4 20-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – implementing neural tasks for week 1, with
networks initiative and attention to
detail.

5 21-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – comparing models and tasks for week 1, with
selecting best performer initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 21

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 2
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 24-02- 10:00 05:00 Worked on handwritten digits Consistently delivered
2025 AM PM recognition – exploring MNIST dataset tasks for week 2, with
and loading into notebook initiative and attention to
detail.

2 25-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – preprocessing images tasks for week 2, with
and visualizing data initiative and attention to
detail.

3 26-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – applying basic classifiers tasks for week 2, with
like KNN and SVM initiative and attention to
detail.

4 27-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – implementing neural tasks for week 2, with
networks initiative and attention to
detail.

5 28-02- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – comparing models and tasks for week 2, with
selecting best performer initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 22

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 3
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 03-03- 10:00 05:00 Worked on handwritten digits Consistently delivered
2025 AM PM recognition – exploring MNIST dataset tasks for week 3, with
and loading into notebook initiative and attention to
detail.

2 04-03- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – preprocessing images tasks for week 3, with
and visualizing data initiative and attention to
detail.

3 05-03- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – applying basic classifiers tasks for week 3, with
like KNN and SVM initiative and attention to
detail.

4 06-03- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – implementing neural tasks for week 3, with
networks initiative and attention to
detail.

5 07-03- 10:00 05:00 Worked on handwritten digits Consistently delivered

2025 AM PM recognition – comparing models and tasks for week 3, with
selecting best performer initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 23

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 4
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 10-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered
2025 AM PM exploring FIFA player dataset tasks for week 4, with
initiative and attention to
detail.

2 11-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM analyzing player attributes and skills tasks for week 4, with
initiative and attention to
detail.

3 12-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM clustering based on roles (striker, tasks for week 4, with
winger, etc.) initiative and attention to
detail.

4 13-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM visualizing age vs. ratings and salary tasks for week 4, with
trends initiative and attention to
detail.

5 14-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM drafting analysis report and tasks for week 4, with
interpreting data initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 24

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 5
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 17-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered
2025 AM PM exploring FIFA player dataset tasks for week 5, with
initiative and attention to
detail.

2 18-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM analyzing player attributes and skills tasks for week 5, with
initiative and attention to
detail.

3 19-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM clustering based on roles (striker, tasks for week 5, with
winger, etc.) initiative and attention to
detail.

4 20-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM visualizing age vs. ratings and salary tasks for week 5, with
trends initiative and attention to
detail.

5 21-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM drafting analysis report and tasks for week 5, with
interpreting data initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 25

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 6
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 24-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered
2025 AM PM exploring FIFA player dataset tasks for week 6, with
initiative and attention to
detail.

2 25-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM analyzing player attributes and skills tasks for week 6, with
initiative and attention to
detail.

3 26-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM clustering based on roles (striker, tasks for week 6, with
winger, etc.) initiative and attention to
detail.

4 27-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM visualizing age vs. ratings and salary tasks for week 6, with
trends initiative and attention to
detail.

5 28-03- 10:00 05:00 Worked on fifa 20 player analysis – Consistently delivered

2025 AM PM drafting analysis report and tasks for week 6, with
interpreting data initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 26

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 7
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 31-03- 10:00 05:00 Worked on pubg match winner Consistently delivered
2025 AM PM prediction – preprocessing and feature tasks for week 7, with
engineering initiative and attention to
detail.

2 01-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – training predictive models tasks for week 7, with
for match outcome initiative and attention to
detail.

3 02-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – analyzing winPlacePerc tasks for week 7, with
and top features initiative and attention to
detail.

4 03-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – visualizing player tasks for week 7, with
movements and statistics initiative and attention to
detail.

5 04-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – report generation and tasks for week 7, with
model comparison initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 27

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 8
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 07-04- 10:00 05:00 Worked on pubg match winner Consistently delivered
2025 AM PM prediction – preprocessing and feature tasks for week 8, with
engineering initiative and attention to
detail.

2 08-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – training predictive models tasks for week 8, with
for match outcome initiative and attention to
detail.

3 09-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – analyzing winPlacePerc tasks for week 8, with
and top features initiative and attention to
detail.

4 10-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – visualizing player tasks for week 8, with
movements and statistics initiative and attention to
detail.

5 11-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – report generation and tasks for week 8, with
model comparison initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 28

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 9
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 14-04- 10:00 05:00 Worked on pubg match winner Consistently delivered
2025 AM PM prediction – preprocessing and feature tasks for week 9, with
engineering initiative and attention to
detail.

2 15-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – training predictive models tasks for week 9, with
for match outcome initiative and attention to
detail.

3 16-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – analyzing winPlacePerc tasks for week 9, with
and top features initiative and attention to
detail.

4 17-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – visualizing player tasks for week 9, with
movements and statistics initiative and attention to
detail.

5 18-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – report generation and tasks for week 9, with
model comparison initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 29

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 10
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 21-04- 10:00 05:00 Worked on pubg match winner Consistently delivered
2025 AM PM prediction – preprocessing and feature tasks for week 10, with
engineering initiative and attention to
detail.

2 22-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – training predictive models tasks for week 10, with
for match outcome initiative and attention to
detail.

3 23-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – analyzing winPlacePerc tasks for week 10, with
and top features initiative and attention to
detail.

4 24-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – visualizing player tasks for week 10, with
movements and statistics initiative and attention to
detail.

5 25-04- 10:00 05:00 Worked on pubg match winner Consistently delivered

2025 AM PM prediction – report generation and tasks for week 10, with
model comparison initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 30

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 11
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 28-04- 10:00 05:00 Worked on heart disease prediction – Consistently delivered
2025 AM PM understanding heart disease dataset tasks for week 11, with
initiative and attention to
detail.

2 29-04- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM training classification models tasks for week 11, with
(LogReg, SVM) initiative and attention to
detail.

3 30-04- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM evaluating results using metrics like tasks for week 11, with
ROC-AUC initiative and attention to
detail.

4 01-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM deriving health insights and tasks for week 11, with
recommendations initiative and attention to
detail.

5 02-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM finalizing results and visualizations tasks for week 11, with
initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 31

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 12
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 05-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered
2025 AM PM understanding heart disease dataset tasks for week 12, with
initiative and attention to
detail.

2 06-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM training classification models tasks for week 12, with
(LogReg, SVM) initiative and attention to
detail.

3 07-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM evaluating results using metrics like tasks for week 12, with
ROC-AUC initiative and attention to
detail.

4 08-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM deriving health insights and tasks for week 12, with
recommendations initiative and attention to
detail.

5 09-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM finalizing results and visualizations tasks for week 12, with
initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 32

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 13
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 12-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered
2025 AM PM understanding heart disease dataset tasks for week 13, with
initiative and attention to
detail.

2 13-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM training classification models tasks for week 13, with
(LogReg, SVM) initiative and attention to
detail.

3 14-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM evaluating results using metrics like tasks for week 13, with
ROC-AUC initiative and attention to
detail.

4 15-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM deriving health insights and tasks for week 13, with
recommendations initiative and attention to
detail.

5 16-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM finalizing results and visualizations tasks for week 13, with
initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 33

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 14
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 19-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered
2025 AM PM understanding heart disease dataset tasks for week 14, with
initiative and attention to
detail.

2 20-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM training classification models tasks for week 14, with
(LogReg, SVM) initiative and attention to
detail.

3 21-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM evaluating results using metrics like tasks for week 14, with
ROC-AUC initiative and attention to
detail.

4 22-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM deriving health insights and tasks for week 14, with
recommendations initiative and attention to
detail.

5 23-05- 10:00 05:00 Worked on heart disease prediction – Consistently delivered

2025 AM PM finalizing results and visualizations tasks for week 14, with
initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 34

Internship Report

STUDENT’S DAY-WISE DAIRY

Week - 15
Day Date Login Logout Topics Learnt Remarks by Supervisor
1 26-05- 10:00 05:00 Worked on final documentation and Consistently delivered
2025 AM PM submission – final review of all tasks for week 15, with
capstone notebooks initiative and attention to
detail.

2 27-05- 10:00 05:00 Worked on final documentation and Consistently delivered

2025 AM PM submission – editing markdown cells tasks for week 15, with
and plots initiative and attention to
detail.

3 28-05- 10:00 05:00 Worked on final documentation and Consistently delivered

2025 AM PM submission – writing conclusion tasks for week 15, with
reports for each project initiative and attention to
detail.

4 29-05- 10:00 05:00 Worked on final documentation and Consistently delivered

2025 AM PM submission – formatting output and tasks for week 15, with
documentation initiative and attention to
detail.

5 30-05- 10:00 05:00 Worked on final documentation and Consistently delivered

2025 AM PM submission – final submission of tasks for week 15, with
project notebook initiative and attention to
detail.

Sameera Sultana Internal Guide Internship Coordinator

Industry Supervisor

Dept. of AI&ML, GNDECB P a g e | 35

Internship Report

4.2 Glimpse of Project Execution

Fig 4.2.1 Our Instructors Reviewing our code

Fig 4.2.2 Our Team Collaborating For Project

Dept. of AI&ML, GNDECB P a g e | 36

Internship Report

CHAPTER 5:

MINI PROJECTS

This chapter provides detailed descriptions of the four mini projects carried out during the
internship. Each project was a substantial practical exercise in applying data science and
machine learning techniques to a specific problem domain. The projects are presented in
the order they were undertaken:

 Handwritten Digits Recognition: A computer vision task to classify handwritten

images of digits (0-9) using machine learning.
 FIFA 20 Player Analysis: A data analysis and clustering task on a dataset of
football players to extract insights and group similar players.
 PUBG Match Winner Prediction: A predictive modeling task to estimate the
probability of winning a battle royale game based on player statistics.
 Heart Disease Prediction: A healthcare analytics task to predict the presence of
heart disease in patients from clinical data and suggest preventative measures.

For each project, the problem statement is outlined, followed by the approach and tools
used, and finally the outcomes and insights gained. The narrative also touches on any
significant challenges faced and how they were addressed. These projects illustrate the
application of theoretical knowledge in real scenarios and demonstrate the learning
progression of the intern.

5.1 Handwritten Digits Recognition

Problem Statement: The goal of this project was to develop a model that can accurately
recognize handwritten digits (0 through 9) from images. This is a classic problem in
computer vision and machine learning, often approached using the MNIST dataset, which
contains thousands of 28x28 pixel grayscale images of handwritten digits. The task was
two-fold: first, perform an exploratory data analysis on the digit image data, and second,
build and compare multiple classification models to determine which provides the best

Dept. of AI&ML, GNDECB P a g e | 37

Internship Report

accuracy in classifying the images. Ultimately, the best-performing model would be

recommended for use in a production scenario where such digit recognition might be
needed (for example, automated reading of handwritten forms).

Approach: The intern began by exploring the MNIST dataset. Summary statistics like the
count of images per digit (which is typically uniform in MNIST) were checked, and sample
images of each digit were visualized to understand the variation in handwriting styles. Each
image in the dataset is essentially 784 features (pixels) when flattened, with each pixel
intensity as a value. Recognizing that dealing with all pixels directly can be
computationally intensive, the intern ensured to implement efficient data handling using
NumPy arrays. No significant data cleaning was required as MNIST is a well-prepared
dataset, but normalization of pixel values (scaling from 0-255 to 0-1) was done to aid
certain algorithms.

For the modeling part, three types of classifiers were built and evaluated:

1. K-Nearest Neighbors (KNN): As a simple baseline, a KNN classifier (with a

suitable choice of k, e.g. 3 or 5) was used. KNN predicts the class of a new image
by looking at the classes of the closest images in the training set (in terms of pixel
distance). While easy to implement, KNN can be slow on large datasets and doesn’t
create an explicit “model” beyond the stored data.
2. Support Vector Machine (SVM): An SVM with a non-linear kernel (such as
RBF) was applied to find an optimal boundary between digit classes in the high-
dimensional pixel space. SVMs are known to perform well on recognition tasks by
creating complex decision boundaries. The intern used scikit-learn’s SVM
implementation and experimented with parameters like the kernel type and
regularization parameter C. Training an SVM on all 60,000 MNIST training images
is computationally heavy, but a subset was used initially to tune parameters, and
then a full training was done for the final model.
3. Neural Network: Given that digit recognition is famously solved to high accuracy
by neural networks, the intern also implemented a simple feed-forward neural
network (multi-layer perceptron) using Keras. The network architecture consisted

Dept. of AI&ML, GNDECB P a g e | 38

Internship Report

of an input layer (784 nodes), one or two hidden layers with a reasonable number
of neurons (e.g., 128 neurons each) using ReLU activation, and an output layer of
10 neurons (one for each digit class) using softmax activation. The network was
trained for several epochs on the training data, using a portion of it as a validation
set to monitor performance.

Throughout these experiments, the intern used the same training and test splits for fair
comparison, and tracked the accuracy of each model on the test set. Cross-validation was
also used for the non-neural models to ensure results were consistent across different
subsets of data.

Results: All models achieved a high degree of accuracy, which is expected given that
MNIST is a well-researched problem where even simple models perform reasonably. The
KNN model, while straightforward, achieved a decent accuracy (around 95%) but was the
slowest at prediction time, since it had to compute distances to many training points for
each new image. The SVM model improved on accuracy, reaching around 97-98% test
accuracy, and proved to be a strong candidate. The neural network model also reached
about 98% accuracy on the test set after tuning (for instance, training for 20 epochs, using
an appropriate optimizer like Adam, and perhaps adding a dropout layer to avoid
overfitting). In the end, the neural network slightly outperformed the SVM by a small
margin in accuracy, and importantly, it produced very fast predictions once trained (since
classification is just a series of matrix multiplications). Therefore, the neural network was
suggested as the best model for deployment, given its combination of accuracy and
efficiency.

Model Comparison and Insights: The project required not just finding the best model but
also documenting the comparison. The intern noted that the SVM was effective but training
it on the full dataset took significant time and memory. The KNN was simple but not
scalable to real-time use with large datasets. The neural network required tuning of
hyperparameters like the number of neurons and epochs, but once tuned, generalized very
well to unseen data. One interesting observation was how the models made mistakes: by
examining some of the images that were misclassified by each model, the intern found that

Dept. of AI&ML, GNDECB P a g e | 39

Internship Report

certain digits that were written in an unusual way (for example, a sloppy "5" that looked
like a "6") confused even the best model. This highlighted the inherent challenge in
handwriting recognition and the importance of large diverse training data.

Challenges: A key challenge encountered was choosing the right hyperparameters without
overfitting. For instance, if the neural network was made too large (too many hidden
neurons or layers), it could memorize the training images and perform slightly worse on
test images. The intern addressed this by using validation data to guide model complexity
and by employing regularization techniques (such as dropout). Another challenge was
computational resources: training the neural network and SVM on a standard laptop was
time-consuming, so the intern utilized minibatch training for the neural net and used only
a subset of data for initial SVM tuning. Documentation of these challenges and how they
were mitigated was included in the project report as required.

In conclusion, the Handwritten Digits Recognition project was successful, achieving near
state-of-the-art performance on the MNIST dataset. It provided the intern with hands-on
experience in computer vision and classification modeling. The project demonstrated how
different algorithms can be applied to the same problem and reinforced the importance of
model evaluation and selection based on both performance and practicality.

Dept. of AI&ML, GNDECB P a g e | 40

Internship Report

5.1.1 Outputs

Fig [Link] Comparison Samples

Fig [Link] Comparison Table

Dept. of AI&ML, GNDECB P a g e | 41

Internship Report

Fig [Link] Data Analysis

Fig [Link] Conclusion

Dept. of AI&ML, GNDECB P a g e | 42

Internship Report

5.2 FIFA 20 Player Analysis

Problem Statement: The FIFA 20 Player Analysis project was centered on deriving
insights from a comprehensive dataset of football (soccer) players. The dataset spanned
players featured in EA Sports’ FIFA 20 game, including various attributes such as age,
nationality, overall skill rating, potential rating, wage, value, and a host of skill metrics
(like speed, shooting, passing, defending, etc.). The project had multiple objectives:

 Perform a thorough data analysis to discover key trends (for example, identify
which countries produce the most professional players in the dataset).
 Investigate the relationship between player attributes and age (to determine the
typical peak age of player performance/improvement).
 Examine the salary (or value) differences among offensive player positions (strikers
vs. right-wingers vs. left-wingers) to see which position tends to be valued the most.
 Apply clustering to group players based on their attributes, in order to find natural
groupings (which could correspond to player types or positions). This project was
less about predictive modeling and more about exploratory analysis and
unsupervised learning (clustering) to make sense of a real-world sports data set.

Approach: The intern’s approach began with data cleaning and exploration. The FIFA
dataset, usually provided as a CSV, was loaded into a Pandas DataFrame. The intern
checked for missing values or anomalies; for instance, some player entries might have
missing wage information or could have a value of 0 for certain skills if not applicable. In
such cases, decisions were made on whether to fill, drop, or ignore those fields. Units were
standardized (heights might be given in inches in the raw data but were converted to
centimeters for consistency, as noted in the dataset documentation).

For the analysis part:

 To find the top countries producing players, the intern grouped the data by
nationality and counted the number of players per country. This resulted in a
ranking of countries. A bar chart was created to visualize the top 10 countries. As
expected, countries with large football talent pools (like England, Spain,

Dept. of AI&ML, GNDECB P a g e | 43

Internship Report

Germany, Argentina, Brazil, France, etc.) featured in the top list, confirming
real-world expectations that those nations produce many professional players.
 Investigating player improvement with age involved analyzing the overall rating
vs. age. The intern plotted age on the x-axis and overall rating on the y-axis for all
players. A trend emerged: younger players (in their teens and early 20s) usually
have lower overall ratings (since they are still developing), and ratings improve as
age increases, peaking in the late 20s. After around 30 years of age, many players’
ratings begin to decline. To pinpoint an approximate “peak age,” the intern
computed the average overall rating for players in each age and observed the
maximum. The analysis suggested that players tend to reach their peak performance
(highest overall ability) around 27-29 years old, after which the average rating
either plateaus or decreases. This aligns with common sports science knowledge
that late 20s are a footballer’s prime years.
 For the salary comparison among offensive positions, the dataset had information
on player positions (often multiple positions, but the primary position was
considered). The intern filtered players by position: Strikers (ST), Left Wingers
(LW), and Right Wingers (RW). The average wage of players in each of these
categories was calculated, or alternately, a distribution of wages for each category
was plotted. The insight drawn was that strikers often command high wages,
frequently being the focal point of a team’s attack and often in high demand.
However, exceptional wingers (like world-class left or right wingers) also have
very high wages, so the data might show significant overlap. In many team settings,
the difference was not huge, but if averaged across the dataset, strikers showed a
slightly higher mean wage. The intern described this finding with the caveat that
wage can be influenced by many factors (like overall rating and club wealth)
beyond just the position.
 Clustering analysis: The intern selected a subset of attributes to cluster players.
Using too many attributes (FIFA has dozens per player) could complicate
clustering, so focus was on key performance attributes that define play style, such
as pace (speed), shooting, passing, dribbling, defending, and physical abilities.
These attributes were normalized (to ensure, for example, that the scale of shooting,

Dept. of AI&ML, GNDECB P a g e | 44

Internship Report

rated 1-100, didn’t dominate another attribute). A K-Means clustering algorithm

was applied, with the intern experimenting with different numbers of clusters (k).
Using the elbow method (plotting the within-cluster sum of squares for increasing
k), an optimal k (for example, k=4 or 5) was chosen where adding another cluster
gave diminishing returns. After clustering, the intern interpreted each cluster by
looking at the centroid values of attributes and the players in each cluster. An
example interpretation might be:
o Cluster 1: High pace and dribbling, moderate shooting – these turned out to
be winger-type players who are very fast and good at dribbling.
o Cluster 2: High shooting and moderate pace – these were often strikers who
focus on scoring.
o Cluster 3: High defending and physicality – mostly defenders.
o Cluster 4: Balanced attributes or goalkeepers (depending on if goalkeepers
were included or separated since they have distinct attributes). The
clustering gave a data-driven confirmation of player types corresponding
largely to their positions, but also revealed groups such as versatile
midfielders who had a balance of passing and dribbling.

Outcomes and Insights: The analysis yielded several insights:

 Country-wise Talent Production: A list of top 10 countries by number of players

was compiled. For example, the analysis might list: 1) England, 2) Germany, 3)
Spain, 4) Argentina, 5) France, 6) Brazil, 7) Italy, 8) Colombia, 9) Japan, 10)
Netherlands (hypothetical order). This shows European and South American
nations dominate, with some surprises like Japan due to the large dataset including
many players.
 Player Peak Age: It was clearly observed that player performance improves with
age until around the late 20s. After ~30, performance declines, as evidenced by the
decreasing overall ratings for older players. Thus, teams might prefer players in the
24-29 age bracket for peak performance, whereas very young players are seen as
investments for the future (high potential but currently lower overall ratings).

Dept. of AI&ML, GNDECB P a g e | 45

Internship Report

 Offensive Position Salaries: The wage analysis suggested that while strikers
generally earn the most (given their crucial role in scoring), the difference between
top strikers and top wingers was not vast. It was noted that market value is often
more tied to overall skill and star power than just position – for instance, a superstar
right winger could earn more than an average striker. The intern concluded that
among attacking roles, strikers slightly edge out others in average value, but each
position has elite players with very high wages.
 Player Clusters: The clustering provided a neat segmentation of players. It
effectively grouped players by playing style and strengths. This could be useful for
things like identifying what type of player a club needs (for example, if a club needs
a fast dribbler, they look at players from the cluster characterized by high pace and
dribbling). The intern provided example players from each cluster to illustrate, like
“Cluster 1 (speedy wingers) included players such as X and Y, who are known for
their acceleration and crossing ability,” etc. The clusters mostly aligned with known
positions but also highlighted outliers (like an unusually fast defender might appear
in a cluster with midfielders).

Challenges: Working with the FIFA dataset had its challenges. One was the sheer number
of features – with over 80 attributes, dimensionality reduction or careful feature selection
was needed for clustering to be meaningful. The intern overcame this by focusing on the
most relevant features for the questions at hand. Another challenge was that some data
points could skew analysis (for example, if a particular country had an excessively large
number of low-tier players, it might top the count but those players might be of lower
quality; hence, the intern also looked at average player ratings by country to complement
the count).

In summary, the FIFA 20 Player Analysis project allowed the intern to practice data
analysis skills on a rich dataset and extract insights akin to what a sports analyst or football
club might be interested in. The project did not result in a single predictive model, but
rather a collection of findings and visualizations that tell a story about the data. These
findings were well-documented, and the intern gained experience in handling real-world
data complexities, as well as presenting analytical results clearly.

Dept. of AI&ML, GNDECB P a g e | 46

Internship Report

5.2.1 Outputs FIFA 20 Player Analysis

Fig [Link] Data Set Analysis

Fig [Link] Data Preprocessing

Fig [Link] Data Cleaning

Dept. of AI&ML, GNDECB P a g e | 47

Internship Report

Fig [Link] Data Clustering using Seaborn

Fig [Link] age Distribution using Seaborn

Dept. of AI&ML, GNDECB P a g e | 48

Internship Report

Fig [Link] Height Distribution using Seaborn

Fig [Link] Weight Distribution using Seaborn

Dept. of AI&ML, GNDECB P a g e | 49

Internship Report

Fig [Link] Optimal k Graph

Fig [Link]

Fig [Link] Top 10 Countries by Number of Players

Dept. of AI&ML, GNDECB P a g e | 50

Internship Report

Fig [Link] Overall Rating vs Age of Player

Dept. of AI&ML, GNDECB P a g e | 51

Internship Report

5.3 PUBG Match Winner Prediction

Problem Statement: The PUBG Match Winner Prediction project was focused on
analyzing game data from the popular online multiplayer game PlayerUnknown’s
Battlegrounds (PUBG). The primary objective was to create a model that predicts a
player’s (or team’s) likelihood of winning a match (often represented by a variable like
“winPlacePerc,” which is a normalized ranking outcome of the match). In addition, the
project sought to determine which in-game factors most influence the probability of
winning. The dataset contained extensive match statistics for players or teams in many
PUBG matches, including features such as number of kills, damage dealt to opponents,
distance traveled on foot or by vehicle, number of survival items used (heals, boosts), and
more. By analyzing and modeling this data, the intern aimed to capture how gameplay
metrics translate into winning chances, essentially teaching a model to predict match
outcomes from player performance stats.

Approach: The intern’s approach combined exploratory data analysis with predictive
modeling:

 Exploratory Analysis: Initially, the intern examined distributions of key variables.

For example, most players get 0 kills in a match (since only one team wins, many
players die with few or no kills), and only a tiny fraction of players get high kill
counts (10+ kills). The distribution of “winPlacePerc” (the target, ranging from 0
to 1 where 1 means first place) was also reviewed; it is typically uniform or slightly
skewed depending on ranking definition. The intern plotted how some features
correlate with winPlacePerc. It was apparent that players who survive longer
(reflected by greater distance traveled or more healing items used) tend to have
higher winPlacePerc. Similarly, higher kills and damage dealt strongly indicate a
better finish position. These observations grounded the intern’s expectations for the
model (e.g., any good model should assign higher win probabilities to players with
many kills, all else equal).
 Feature Engineering: Before feeding the data to models, some new features were
engineered. The intern combined certain features to make them more meaningful;

Dept. of AI&ML, GNDECB P a g e | 52

Internship Report

for example, summing up distances (total_distance = walkDistance + rideDistance

+ swimDistance) gave a single measure of how far a player moved, which correlates
with survival time (since moving longer distances generally means staying alive
longer). Another engineered feature was a kill-to-distance ratio as a proxy for
aggressiveness (players with high kills but low distance might have dropped into a
hot zone and fought intensely, for instance). Also, categorical features like match
type (solo, duo, squad) were considered: the intern filtered or stratified the data by
match type if needed because dynamics differ (in a squad match, one player’s stats
might not tell the whole team outcome story).
 Model Selection: For prediction, the intern primarily treated it as a regression
problem to predict the win placement percentage as a continuous outcome. A few
algorithms were tried:
o Linear Regression: as a baseline to see how a simple linear combination
of features fares. This provided a benchmark but likely underfit given the
complex relationships.
o Decision Tree/Random Forest Regressor: Decision trees can capture non-
linear interactions and are intuitive (they might split on “kills >= 1” early,
separating those who at least got a kill, etc.). A Random Forest, being an
ensemble of many trees, was used to improve accuracy and generalization.
The intern trained a Random Forest on a subset of features and used out-of-
bag evaluation or a validation set to gauge its performance.
o Gradient Boosting (e.g., XGBoost or LightGBM): Given the tabular data
nature and the importance of squeezing out predictive performance, the
intern also tried a gradient boosted trees model, which often yields high
accuracy on structured data. This model iteratively builds an ensemble,
focusing on reducing error, and can handle large datasets efficiently.
 Model Training and Validation: The dataset was large, so training was done on
a sample or using efficient algorithms. The intern split the data into training and
validation sets. Cross-validation was used if feasible (though on a very large
dataset, a single hold-out might have sufficed due to computational constraints).
The performance metric for regression was chosen as Mean Absolute Error

Dept. of AI&ML, GNDECB P a g e | 53

Internship Report

(MAE) or Root Mean Squared Error (RMSE), as predicting the exact rank can
be noisy, but minimizing average error is a good target. If the intern also attempted
classification (like predicting top-10 placement vs not), then accuracy and F1-score
were looked at.
 Feature Importance Analysis: After training the best model (Random Forest or
XGBoost), the intern extracted feature importances. This gave a ranking of which
features contributed most to the prediction. Typically, one would expect features
like kills, walkDistance, and damageDealt to be very important. Also, heals and
boosts (items used) often indicate a long survival and thus impact outcome. The
intern documented these, as they directly answer the “important factors” part of the
problem.

Results: The model achieved a reasonable level of accuracy in predicting win outcomes.
For example, the Random Forest model might have an RMSE of around 0.05-0.06 in
predicting winPlacePerc (on a 0-1 scale), which translates to an average error of about 5-
6% in rank prediction, or an MAE of say 0.04 (4%). In more intuitive terms, if the true
placement of a player is 1st (100% winPlacePerc), the model might predict something like
0.95 (95%), or if a player placed 50th percentile, the model might predict 0.46 or 0.55, etc.
This is quite good given the inherent randomness (a player with good stats could still lose
due to a single mistake at the end, etc.). If classification was attempted for, say, “will this
team win the match (yes/no)”, the accuracy might have been in the high 80s to 90% range,
but note that in a large match of 100 players, predicting “no” for everyone yields 99%
accuracy trivially (since only one wins), so classification would need careful interpretation
(hence regression is more informative).

The feature importance ranked the factors as anticipated:

 Kills: This was one of the top predictors. The more kills a player (or team) has, the
more likely they survived longer and eliminated competition, which correlates with
winning. The model recognized this, assigning a high importance to the kills
feature.

Dept. of AI&ML, GNDECB P a g e | 54

Internship Report

 Distance Traveled (WalkDistance in particular): This feature also ranked very

high. It essentially measures how far a player moved, indirectly measuring survival
time and activity. Someone who only moved 100 meters likely died very early,
whereas someone with 4000 meters probably was alive till late game. The model
heavily used this in predictions.
 DamageDealt: Total damage dealt to opponents was another key factor. It captures
combat engagement. Even if not every damage results in a kill, doing a lot of
damage means a player was actively fighting and likely surviving those fights.
 Heals and Boosts: The usage of healing and boost items was also important.
Players who survive long enough to use many healing kits or energy drinks are
often in the final circles of the game, hence more likely to be winners. The model
used these as indicators of longevity in the match.
 Survival Metrics: Other metrics like weaponsAcquired (players picking up many
weapons might have looted extensively, meaning they lived longer) or killStreaks
(multiple kills in a short time) provided additional signals. Less influential features
might include things like rideDistance (useful but maybe redundant if walkDistance
is already included), or assists (helpful but not as impactful as kills in a solo
context).

The intern also highlighted an interesting insight: while kills and damage are critical, a
player doesn’t need the absolute highest kills to win—strategic play and survival reflected
by distance and healing can sometimes compensate. For example, a player with moderate
kills but who stayed safe and healed at the right times could still win, which the model
accounts for via those features.

Challenges: One major challenge was the size of the data. PUBG data sets can be millions
of rows. The intern had to sample data or use computational resources efficiently (like
using vectorized operations and perhaps leveraging the GPU for XGBoost if available).
Another challenge was multicollinearity between features; for instance, killPoints,
rankPoints (legacy ranking metrics in some PUBG data) correlate with kills and placement.
The intern decided to drop or ignore some of these to not leak information (some datasets
have a variable winPoints which correlates directly with winPlacePerc, which if used

Dept. of AI&ML, GNDECB P a g e | 55

Internship Report

would trivially let the model cheat; so it was excluded). Ensuring a proper train-test split
by match (to avoid data leak where one team’s info could influence prediction of another
in the same match if not careful) was also handled. The intern learned the importance of
domain context—realizing, for example, that in a squad match, individual stats alone might
be insufficient and one might need team aggregates. For simplicity, if the project kept to
solo match data, this issue is minimized.

In conclusion, the PUBG Match Winner Prediction project successfully demonstrated how
data science can be applied to gaming analytics. The intern built a model that can
reasonably predict outcomes and, more importantly, extracted the key factors that drive
success in the game. These factors (staying alive, securing kills, and continuously looting
and moving) resonate with actual player strategies, thereby validating the model’s findings.
The project enhanced the intern’s skills in handling large datasets, feature engineering, and
interpreting machine learning models in terms of real-world behavior.

5.3.1 Outputs

Fig [Link] Data Pre Processing

Dept. of AI&ML, GNDECB P a g e | 56

Internship Report

Fig [Link] Distribution of win Place Percentage

Fig [Link] Correlation Matrix of Numerical Features

Dept. of AI&ML, GNDECB P a g e | 57

Internship Report

Fig [Link] Conclusion

Dept. of AI&ML, GNDECB P a g e | 58

Internship Report

5.4 Heart Disease Prediction

Problem Statement: The Heart Disease Prediction project dealt with a medical dataset
and aimed to create a machine learning model to predict whether a person has heart disease
based on various health measurements. The dataset (a heart disease dataset, possibly
derived from sources like the UCI Heart Disease dataset or a Kaggle cardiovascular
dataset) included features such as age, sex, chest pain type, resting blood pressure,
cholesterol level, fasting blood sugar, resting ECG results, maximum heart rate achieved,
exercise-induced angina, ST depression (oldpeak) and others, along with a target label
indicating the presence of heart disease. In addition to building a predictive model, the
project required providing suggestions to the hospital on how to act on these predictions to
prevent life-threatening events. In essence, the intern needed to not only classify patients
as at risk or not, but also interpret the model to highlight risk factors and recommend
preventive measures.

Approach: The approach combined supervised learning with domain understanding:

 Data Understanding and Preprocessing: The intern began by reviewing each

feature’s definition (some of which were detailed in the project description). For
example, exercise_induced_angina (yes/no if exercise causes chest pain), oldpeak
(ST depression induced by exercise, indicating possible ischemia), thal (result of a
thallium stress test, which can be normal or show defects). Knowing what each
means helped in anticipating their relationship with heart disease (e.g., exercise-
induced angina is likely a strong indicator of heart problems). The dataset was
checked for missing values or unusual distributions. In many heart datasets, missing
values might appear in features like thal or num_major_vessels due to test results
not available for some patients. The intern applied imputation or removed those
records as appropriate (if few).

Next, categorical features were encoded: for instance, thal categories (“normal”,
“fixed defect”, “reversible defect”) were converted to numeric dummy variables.
The chest pain type (4 categories) might also be one-hot encoded or treated as an

Dept. of AI&ML, GNDECB P a g e | 59

Internship Report

ordinal if the numbers 1-4 indicated increasing severity of pain. Continuous

features like age, blood pressure, cholesterol were left as is or normalized if needed
for certain models. Given many features were on different scales (cholesterol in
mg/dl, max heart rate in bpm, etc.), scaling was considered, particularly for
algorithms like logistic regression or SVM.

 Exploratory Analysis: The intern likely examined how each feature correlates
with the presence of heart disease. For example, they might have observed that
patients with heart disease in the data tend to have higher frequencies of certain
attributes: a higher proportion of them have exercise-induced angina, higher resting
blood pressure on average, lower max heart rate achieved (since heart disease can
limit exercise capacity), etc. Perhaps a quick check: the average age of those with
heart disease might be higher than those without, and the proportion of males might
be higher (as historically some heart datasets have more male patients). This
provided a sanity check that the data is consistent with medical knowledge (e.g.,
older age and certain symptoms correlate with disease).
 Model Development: The intern initially tried a Logistic Regression model,
which is a common choice for binary medical outcomes. Logistic regression gives
a probabilistic output (probability of heart disease) and allows one to gauge
influence of each feature through its coefficients. The model was trained using a
portion of data and evaluated on a validation set. Given a moderate dataset size
(often heart disease datasets have a few hundred rows; some newer ones have
more), cross-validation (like 5-fold CV) was likely used to make the most of the
data and ensure stability of results.

After logistic regression, the intern explored more complex models:

o Decision Tree: A decision tree could find rule-based patterns, like “if
oldpeak > 1.5 and exercise_induced_angina = yes, then disease = yes”. The
intern tuned the tree to avoid overfitting (setting a max depth).
o Random Forest: This ensemble of trees often improves accuracy. The
intern trained a Random Forest classifier and used its out-of-bag error or

Dept. of AI&ML, GNDECB P a g e | 60

Internship Report

validation performance to evaluate it. Random Forests also provide feature

importance which was useful.
o Possibly Support Vector Machine or Naive Bayes: SVM could be tried
but might not add much insight beyond logistic regression for this task, and
Naive Bayes can be a quick baseline assuming feature independence.

Ultimately, the Random Forest or logistic regression likely emerged as top choices
– Random Forest for raw accuracy, logistic for interpretability. Suppose the
Random Forest achieved an accuracy of around 88% on validation, and logistic
regression was around 85%. The intern might choose the Random Forest as the
final model to maximize predictive performance.

 Evaluation: To properly evaluate, the intern looked at the confusion matrix from
the best model: how many patients were correctly identified as having heart disease
vs missed (false negatives) vs false alarms (false positives). In a medical context,
catching as many true cases (high recall/sensitivity) is important, even if it means
some false positives (which can be later tested further by doctors). The precision is
also considered because too many false positives could overwhelm resources or
cause unnecessary alarm. The intern reported metrics like Accuracy, Precision,
Recall, and F1-score for the positive class (heart disease). For example, the model
might have a recall of 90% (meaning it catches 90% of true heart disease cases) and
a precision of 85% (meaning 85% of those it flags as high risk indeed have the
disease).
 Insights and Recommendations: Beyond the numbers, the intern interpreted the
model. Using the logistic regression coefficients or Random Forest feature
importances, the significant predictors were identified. Common results: oldpeak,
thal(defect type), max_heart_rate_achieved, exercise_induced_angina,
chest_pain_type, and num_major_vessels are usually top indicators. For instance,
the model might clearly show that having exercise-induced angina and ST
depression (oldpeak) greatly increase the probability of heart disease. High
cholesterol and high resting blood pressure are also contributors but in some

Dept. of AI&ML, GNDECB P a g e | 61

Internship Report

datasets they are not as statistically significant as the exercise and ECG-related
variables. The intern highlighted that:
o Exercise-induced angina (angina during exercise): Strong predictor –
patients experiencing this have a much higher risk of heart disease.
o Oldpeak (ST depression): Every unit increase in oldpeak significantly
raises the odds of heart disease, indicating stress-induced ischemia.
o Chest pain type: Certain types of chest pain (especially typical angina pain)
were strongly associated with heart condition.
o Max heart rate achieved: Lower values (i.e., inability to reach high heart
rates during exercise) often corresponded with presence of heart disease.
o Major vessels colored (from angiography): The more vessels that showed
blockage, the higher the risk.

These align with medical intuition and thus add credibility to the model’s decisions.

Using these insights, the intern formulated recommendations. For example:

o Patients who are older and exhibit multiple risk factors (like high blood
pressure, high cholesterol, abnormal ECG results such as ST depression,
and exercise-induced angina) should be prioritized for further
cardiovascular evaluation (such as a stress test or angiogram if not already
done) and preventive treatment.
o The hospital could integrate the model’s output into their routine check-up
process. If a patient’s data is fed into the model and the predicted risk is
above a certain threshold, an alert could be generated for doctors to conduct
more thorough examinations for heart disease.
o Lifestyle modifications should be suggested for at-risk patients: e.g.,
encourage patients with high cholesterol and high blood pressure (even if
no current heart disease) to adopt diet changes, exercise (as tolerated), and
possibly medication (like statins or antihypertensives) to mitigate these risk
factors. The model in identifying these factors reinforces their importance.

Dept. of AI&ML, GNDECB P a g e | 62

Internship Report

o Regular screening: For patients flagged by the model (or even those with
borderline risk), schedule regular follow-ups. If the dataset covered some
demographic findings (maybe it showed men over 50 with certain factors
are high risk), then ensuring such demographics are regularly screened
would be a recommendation.

The intern likely phrased these suggestions in a general manner since specific
medical advice would be determined by doctors, but the idea was to show how the
data insights can lead to action: focusing on modifiable risk factors (blood pressure,
cholesterol, blood sugar if diabetic), recommending exercise or smoking cessation
if that data were present (some heart datasets include a smoking indicator), etc.

Results: The predictive model achieved solid performance in identifying heart disease. For
example, on a held-out test set, it might have achieved around 85-90% accuracy, with a
sensitivity around 0.9 (meaning it caught 90% of those with disease) and specificity around
0.8-0.85 (correctly ruling out 80-85% of those without disease). This is within a useful
range for a screening tool (though in practice, medical models may undergo further
validation). The key risk factors identified by the model matched well-known medical
knowledge, giving trust in the model.

The hospital suggestions were formulated based on these results: emphasizing early
detection and risk factor management. For instance, because the data showed that many
heart disease patients had high blood pressure and cholesterol, the intern suggested that
controlling these through lifestyle or medication can reduce future heart disease incidents.
Additionally, since exercise ECG indicators (like oldpeak and exercise angina) were
crucial, the intern suggested that stress testing could be used more frequently in the
hospital’s routine for patients above a certain risk profile.

Challenges: One challenge was dealing with the relatively small size of some heart disease
datasets (some have only ~300 patients). This can make it hard for complex models to
generalize. The intern used cross-validation and possibly combined data from multiple
sources if available to improve robustness. Another challenge is class imbalance if present

Dept. of AI&ML, GNDECB P a g e | 63

Internship Report

(if, say, only 30% had disease); this was handled by ensuring performance metrics beyond
accuracy were used, and possibly using techniques like oversampling the minority class or
adjusting classifier thresholds to improve sensitivity. Medical data also can have
multicollinearity (e.g., age and some other factors might correlate), but tree-based models
handle that inherently, whereas logistic regression required checking VIFs or stepwise
selection. The intern navigated these by focusing on the most informative variables and
verifying the model against known medical criteria.

In conclusion, the Heart Disease Prediction project allowed the intern to work on a socially
impactful problem, applying data science to healthcare. The project reinforced the
importance of interpretability in models used for critical decisions and demonstrated how
data-driven models can align with domain expertise to provide both accurate predictions
and actionable insights. The intern concluded the project by summarizing that machine
learning can indeed assist clinicians by flagging high-risk patients and by highlighting
which factors to pay attention to, thus contributing to better preventive care and resource
allocation in the hospital

Dept. of AI&ML, GNDECB P a g e | 64

Internship Report

5.4.1 Outputs of Heart Disease Prediction

Fig [Link] Data Pre Processing

Fig [Link] Data Pre Processing

Fig [Link] Age Distribution

Dept. of AI&ML, GNDECB P a g e | 65

Internship Report

Fig [Link] Correlation matrix

Fig [Link] ROC Curves of Models

Dept. of AI&ML, GNDECB P a g e | 66

Internship Report

Fig [Link] Random Forest Analysis

Fig [Link] Conclusion

Dept. of AI&ML, GNDECB P a g e | 67

Internship Report

CHAPTER 6:

CONCLUSION

The fifteen-week internship at Rubixe seamlessly connected academic theory with hands-
on experience in AI and data science, allowing me to apply Python-based workflows, from
data cleaning and visualization to model selection and performance tuning, across a range
of real-world challenges. Under structured mentorship, I learned to document reproducible
analyses in Jupyter Notebooks, balance accuracy versus efficiency when choosing
algorithms, and translate domain knowledge into actionable insights. Highlights include:
Handwritten Digit Recognition, Football Player Analytics, PUBG Winner Prediction,
Heart Disease Risk Modelling Through these projects, I strengthened my problem-solving
adaptability, project-management skills, and confidence in deploying data-driven
solutions—laying a solid foundation for a future career in AI and Data Science.

Dept. of AI&ML, GNDECB P a g e | 68