Binary Classification - Do an exploratory analysis of the dataset provided, decide on feature selection, preprocessing before training a model to classify as class ‘0’ or class ‘1’.

1. Upload the assignment to github/gitlab (Share the repository link if it’s public or send invite if private) or mail the assignment
2. Split the training data in training and validation set with ratio of 4:1 to evaluate performance of the model on validation set.

The dataset doesn't have the named features. Hence, we can't analyze as per the business / domain.

A high level descriptive analysis of the training dataset was carried out to understand the data. Then, the training dataset was split into two parts - training and validation (4:1 ratio) so that there was no training-validation data leakage in the process and finally, an estimate for the model performance can be calculated. Then, a detailed exploratory data analysis (EDA) of the training dataset was carried out as given below:

• Analysis of missing value - none found
• Analysis of columns with little information
• Analysis of multi-collinearity
• Outlier detection
• Univariate analysis
• Segmented univariate analysis The EDA could provide some idea about the relevant features. However, the features were not finalized yet as there could be non-linear dependencies.
  
  Outcome of EDA: list of redundant features and outliers (to be removed from the training dataset before model training)

First of all, the redundant features and the outliers were removed from the training dataset. Next, a binary classification experiment environment was set to test multiple classification models on this dataset. Next, the best out-of-box classification model was selected as per performance on the training dataset. Then, the model was finetuned and the feature importance was extracted. Then, the top final features set was selected based on the feature importance and shapley value. The selected model was further finetuned after considering the selected features. Finally, I got the fine-tuned classification model, which would be used for model inference.

The validation dataset was used to estimate the model performance. The evaluation metrics - precision, recall, f1-score and AUC - were used to compare and select the model.

The fine-tuned model is used to predict the 'Y' value of the given dataset. The final prediction of the test_set has been provided in the csv file "test_Set_prediction.csv".

Libraries

1. markupsafe == 2.0.1
2. pycaret == 2.3.10
3. seaborn == 0.11.2
4. pandas == 1.3.5
5. scikit-learn == 0.23.2
6. lightgbm == 3.3.2
7. shap == 0.41.0
8. numpy == 1.19.5

• 1_Arya_EDA.ipynb - to be run locally
• 2_Arya_Model_Training_and_Model_Performance_Analysis.ipynb - to be run in google colab
• 3_Arya_Model_Inference.ipynb - to be run in google colab

To run in google colab, we must set our google drive folder. If you get error in importing pycaret.classification, restart the runtime.

1. As the dataset is slightly imbalanced, Hence SMOTE technique was used to balance the dataset.
2. StratifiedKFold strategy was used for cross validation.