Credit Risk Analysis Report

I used lending data to build a machine-learning model that evaluates borrowers and identifies their creditworthiness.

Overview of the Analysis

1. I used the lending_data.csv, which contains loan_size, interest_rate, borrower_income, debt_to_income, num_of_accounts, derogatory_marks, total_debt, and loan_status. The loan_status column contains either 0 or 1, where 0 means that the loan is healthy, and 1 means that the loan is at a high risk of defaulting. I stored the data in a dataframe.

2. To estimate creditworthiness, first, I stored the labels set from the loan_status column in the y variable. Then, I stored the features DataFrame (all the columns except loan_status) in the X variable. I checked the balance of the labels with value_counts. The results showed that, in our dataset, 75036 loans were healthy and 2500 were high-risk.

3. I used the train_test_split module from sklearn to split the data into training and testing variables, these are: X_train, X_test, y_train, and y_test. And I assigned a random_state of 1 to the function to ensure that the train/test split is consistent, the same data points are assigned to the training and testing sets across multiple runs of code.

4. I created a Logistic Regression Model with the original data. I used LogisticRegression(), from sklearn, with a random_state of 1. I fit the model with the training data, X_train and y_train, and predicted on testing data labels with predict() using the testing feature data, X_test, and the fitted model, lr_model.

5. I calculated the accuracy score of the model with balanced_accuracy_score() from sklearn, I used y_test a d testing_prediction to obtain the accuracy.

6. I generated a confusion matrix for the model with confusion_matrix() from sklearn, based on y_test and testing_prediction.

7. I obtained a classification report for the model with classification_report() from sklearn, and I used y_test and testing_prediction.

8. I used RandomOverSampler() from imbalanced-learn to resample the data. I fit the model with the training data, X_train and y_train. I generated resampled data, X_resampled and y_resampled, and used unique() to obtain the count of distinct values in the resampled labels data.

9. Then, I created a Logistic Regression Model with the resampled data, fit the data, and made predictions. Lastly, I obtained the accuracy score, confusion matrix, and classification report of the resampled model.

Results

• Machine Learning Model 1:

  • Model 1 Accuracy: 0.952.
  • Model 1 Precision: for healthy loans the precision is 1.00, for high-risk loans the precision is 0.85.
  • Model 1 Recall: for healthy loans the recall score is 0.99, for high-risk loans the recall score is 0.91.

• Machine Learning Model 2:

  • Model 2 Accuracy: 0.995.
  • Model 2 Precision: for healthy loans the precision is 0.99, for high-risk loans the precision is 0.99.
  • Model 2 Recall: for healthy loans the recall score is 0.99, for high-risk loans the recall score is 0.99.

Summary

• It seems that the logistic regression model with the oversampled data performed better when making predictions.

• Performance depends on the problem we are trying to solve, if we need to predict the healthy loans the logistic regression model with the original data makes better predictions, the classification report generated a precision of 1.00, a recall of 0.99 and an f1-score of 1.00, while with the resampled data the precision was 0.99, the recall was 0.99, and the f1-score was 0.99. However, if we need to predict the high-risk loans the linear regression model with the resampled data does much better, the precision was 0.99, the recall was 0.99, and the f1-score was 0.99, while with the original data, the precision was 0.85, the recall was 0.91, and the f1-score was 0.88.

• It is more important to predict high-risk loans, therefore, I would recommend the second model because it does a much better job of predicting high-risk loans than the first model.