nikatnguyen / credit-risk-classification

Module 12 Report

Overview of the Analysis

This analysis is meant to determine the creditworthiness of borrowers based on loan risk data from a peer-to-peer lending services company. This model was designed to predict high-risk versus healthy loans.

The variables I was trying to predict was the loan_status, which were already encoded into a binary classification of 0 (meaning healthy loan) and 1 (being high-risk).

My process involved basic training and fitting.After loading the data, I separated the data into labels and features, with labels as the "loan_status" column and features being the rest. I then split the data into training and testing datasets using the train_test_split module from sklearn.model_selection. Once they were split, I proceeded to create my logistic regression model using the training data, and I saved predictions on the testing data labels using model.predict. Finally, I generated a confusion matrix for my model and printed a classification report.

Results

• Machine Learning Model 1: Predicting 0 (healthy loans)

  • Model 1 Accuracy score (macro): 0.92
  • The model is 92% accurate
  • Precision: 1.00
  • 100% of the class was found over the whole class
  • Recall: 0.99 -> 99% of the class was correctly classified
  • 99% of the class was correctly classified

• Machine Learning Model 2: Predicted 1 (high-risk loans)

  • Model 2 Accuracy (macro): 0.95
  • The model is 95% accurate
  • Precision: 0.85
  • 85% of the class was found over the whole class
  • Recall: 0.91
  • 91% of the identified class was correctly classified

Summary

The model that seems the best to use would be the first model, which predicted healthy loans. While its macro accuracy score (0.92) was lower than the high-risk accuracy score (0.95), it scored higher for its precision and recall scores, correctly finding and identifying a large majority of healthy loans. For the purpose of trying to identify high-risk loans, neither model would be sufficient to use. The first model predicting healthy loans would be futile to use, and the second model predicting high-risk loans is mostly accurate but requires more work to ensure that it more accurately predicts high-risk loans.