Overview of Credit Card Analysis

In this project, RandomOverSampler and SMOTE algorithms were used to perform oversampling, ClusterCentroids algorithm was used to undersampling, SMOTEENN algorithm was applied as a combinatorial approach of over- and undersampling of credit card credit dataset from LendingClub. Machine learning models - BalancedRandomForestClassifier and EasyEnsembleClassifier were used to predict credit risk.

Results

1. Naive Random Oversampling

2. SMOTE Oversampling

3. Undersampling

4. Combination (Over and Under) Sampling

5. Balanced Random Forest Classifier

6. Easy Ensemble AdaBoost Classifier

Summary

1. Comparing Credit Risk Resampling to Ensemble Techniques, it is clear that higher credit risk prediction accuracy was observed with Easy Ensemble AdaBoost Classifier of 93%. It is recommended that Easy Ensemble AdaBoost Classifier be used to reduce bias in prediction.