Credit Risk Analysis

Preview:

This is a loan prediction risk analysis using credit card dataset from Lending Club, Python and Scikit-Learn libraries to build and evaluate several machine learning models.

Results:

1. Naive Random Oversampling

This model has:

• 65% Accuracy

• 99% Precision

• 69% Recall

• 60% Specificity

• 81% F-score

2. SMOTE Oversampling

This model has:

• 65% accuracy

• 99% precision

• 66% Recall

• 64% Specificity

• 79% F-score

3. Undersampling

This model has:

• 51% Accuracy

• 99% precision

• 40% Recall

• 63% Specificity

• 57% F-Score

4. Over and Under Sampling

This model has:

• 62% Accuracy

• 99% precision

• 54% Recall

• 69% Specificity

• 70% F-Score

5. Balanced Random Forest Classifier

This model has:

• 79% Accuracy

• 99% precision

• 91% Recall

• 67% Specificity

• 95% F-Score

6. Easy Ensemble AdaBoost classifier

This model has:

• 93% Accuracy

• 99% precision

• 94% Recall

• 91% Specificity

• 97% F-Score

Summary:

All machine learning models used in this analysis has the same precision value of 99%. But among all these models used, Easy Ensemble AdaBoost classifier performs the best across all metrics. Therefore, Easy Ensemble AdaBoost is the best classifier in distinguishing between low risk and high risk loans.