#Explanation on Folder Structure:

1. Data : the raw data for which classification is required to be done.
2. notebooks : Feature Engineering and Exploratory Data Analysis
3. models : Trying different Models
4. src : Model Selection

#Feature Engineering and Exploratory Data Analysis:

1. In notebooks folder, there is a file called feature engineering which does following activities:

   • Read data from Training Dataset.
   • Shows datatypes of each feature/variable
   • Count Number of records/row in the dataset
   • differentiate and give count of categorical and numerical features
   • drop categorical feature because of it's uniqueness like an id
   • remove features which are uniqune or have only 1 value like all the values are 0,1, or 100.
   • dealing with missing values using Imputer
   • reduced fatures from 590 to 474 using this process
   • Saved cleaned data into data cleaning.csv for further processing.
   • Check Feature Enginerring.ipynb and Images (Feature Engineering_1.png,Feature Engineering_2.png,Feature Engineering_3.png) for reference.

2. In notebooks folder, there is a file called exploratory data analysis which does following activities:

   • Observations from describe function which explains count, mean, std, min and max values for each feature/variable
   • Shape is 1537 rows and 474 columns
   • Standardization makes each feature's mean to 0 and standard deviation to 1 which helps in calculation when applying different alogrigthms.
   • Train and Test split and standardized scaled Values stored in csv for further processing.
   • K-Means, Heirarchical Clustering and DBScan Cluster alogrithms evaluated for the given data.
   • K-Means and Heirarchical Clustering shows around 2-10 clusters whould be optimium to use for clustering algorithm.
   • DBSCAN algorithm does not show any clusters

3. In src Folder, there is a file called app.py.ipynb where first clustering and then classification algorithm have applied.

   • Libraries load
   • time function defined to calculate the time taken for execution
   • In EDA, it has been noticed DBSCAN was not performing was KMeans and HClust function have been made
   • New class clust have been created with KMeans and HClust Clustering algorithms and then classify function to classify the result
   • clust(load_data).Kmeans(output='add').classify() --> Uses Kmeans clustering with Logistic Regression 'Kmeans' 2453.68 ms Accuracy: 0.9025974025974026 'classify' 9012.03 ms
   • clust(load_data).Kmeans(output='add').classify(SVC()) --> Uses Kmeans clustering with SVC Classification 'Kmeans' 1975.90 ms Accuracy: 0.9329004329004329 'classify' 2802.73 ms
   • clust(load_data).HClust(output='add').classify(SVC()) --> Uses Heirarchical Clustering with SVC Classification Accuracy: 0.9329004329004329 'classify' 2567.15 ms

4. Model Selection

5. Stacking with Classification

   • All the files used here are with .py extension because was required to create Class and import the class to other file and was not able to do it through notebook: This actual test our data and output with Stacking Classification which gives 99% accuracy.
   • onego.py: This stacking technique creates prediction dataset in one go.
   • outoffolds.py: This stacking technique creates prediction dataset by taking the average of the out-of-fold predictors' predictions
   • StackingClassifier.py: This class uses above 2 methods for stacking classification
   • test_stackingclassifier.py: This actual test our data and output with Stacking Classification which gives 99% accuracy.