Description

This is an into example to class imbalance in customer classification problems. It mainly focuses on using preprocessing techniques to solve the problem. In particular it covers:

• Feature Selector
• Dummy Transformer
• Log Transformer
• MinMax Scaler
• KNN Imputer
• SMOTE
• Random UnderSampler

Random Forest and Logistic Regression are used to solve a binary classification problem. The code supports K-fold cross validation and hyper-parameter tuning.

Main Program Execution

Please open a clean terminal window, set your directory to customer-classification/customer_classification and run the file customer_classification.py followed by one of these arguments:

• default: Train, test and evaluate Random Forest and Logistic Regression using a default configuration.
• cv: Perform K-fold cross validation on training data using Random Forest and Logistic Regression.
• tuning: Perform parameter tuning for Random Forest and Logistic Regression while cross-validating the models, it requires setting the hyper-parameters in config/hyper-params.yaml.

Running the code returns a nicely formatted classification report.

Examples:

~/$ python3 customer_classification.py default
~/$ python3 customer_classification.py cv
~/$ python3 customer_classification.py tuning

Unit Testing

The code in test_DataLoaders.py located in folder customer-classification/tests performs a number of tests to the DataLoader class. To run the tests please move to the directory and execute the following command in a clean terminal window:

~/$ python3 test_DataLoaders.py