jncurrea / credit_risk_analysis

Credit Risk Analysis

Overview

Using the credit card credit dataset from LendingClub, a peer-to-peer lending services company, the goal is to oversample the data using the RandomOverSampler and SMOTE algorithms, and undersample the data using the ClusterCentroids algorithm. Then, use a combinatorial approach of over- and undersampling using the SMOTEENN algorithm. Lastly, compare two new machine learning models that reduce bias, BalancedRandomForestClassifier and EasyEnsembleClassifier, to predict credit risk.

Tools

• Jupyter Notebooks
• SciPy
• SKLearn
• imbalanced-learn

Results

Oversampling - Naive Random Oversampling

• Balanced Accuracy Score: 0.65
• Precision Score: 0.01
• Recall Score: 0.63

Oversampling - SMOTE

• Balanced Accuracy Score: 0.65
• Precision Score: 0.01
• Recall Score: 0.64

Undersampling - Cluster Centroids

• Balanced Accuracy Score: 0.50
• Precision Score: 0.01
• Recall Score: 0.59

Combination Sampling - SMOTEENN

• Balanced Accuracy Score: 0.61
• Precision Score: 0.01
• Recall Score: 0.69

Random Forest Classifier

• Balanced Accuracy Score: 0.78
• Precision Score: 0.04
• Recall Score: 0.67

AdaBoost Classifier

• Balanced Accuracy Score: 0.93
• Precision Score: 0.07
• Recall Score: 0.91

Summary

• The AdaBoost Classifier has the highest Balanced Accuracy Score out of all of the classifiers and sampling methods tested in this project. This should be the classifier used to predict risk in this challenge.
• Nonetheless, I would not recommend using any of these classifiers. Even the best one (AdaBoost Classifier) has a low precision for high-risk data points (0.07).