Unit 11 Homework: Risky Business - Rachel Pierce

November 2021

Instructions:

Two Main Tasks:

1. Resampling Technique
2. Ensemble Learning Technique

Please refer to the following documents in GitHub:

• The credit_risk_resampling.ipynb and credit_risk_ensemble.ipynb files (open in Jupyter Lab) for code details.

Part 1

Resampling - Completed.

1. Simple Logistic Regression
   Balanced Accuracy Score = 0.954321

   

2. Oversampling - Naive Random Oversampling
   Balanced Accuracy Score = 0.994828

   

3. SMOTE Oversampling
   Balanced Accuracy Score = 0.994828

   

4. Undersampling
   Balanced Accuracy Score = 0.982881

   

5. Combination (SMOTEENN)
   Balanced Accuracy Score = 0.994748

   

Part 1 Questions:

• Which model had the best balanced accuracy score?
  Answer: Both the oversampling models resulted in the best balanced accuracy scores:
  • Naive Random Oversampling Score = 0.994828
  • SMOTE Oversampling Score = 0.994828
• Which model had the best recall score?
  Answer: They all had 0.99 rec scores.
• Which model had the best geometric mean score?
  Answer: Both the oversampling models and the combination model had 0.99 geo scores.

Part 2

Ensemble Learning - Completed.

1. Balanced Random Forest Classifier Balanced Accuracy Score = 0.5

   

2. Easy Ensemble Classifier Balanced Accuracy Score = 0.926

   

Part 2 Question:

• Which model had the best balanced accuracy score?
  Answer: The Easy Ensemble had the best balanced accuracy score at 0.926.

• Which model had the best recall score?
  Answer: The Balanced Random Forest had the best recall score at 0.99 vs Easy Ensemble at 0.94.

• Which model had the best geometric mean score?
  Answer: The Easy Ensemble had the best geo score at 0.93.

• What are the top three features?
  Answer: The top three features are installment, dti, and loan_amnt as shown in the graph below.