travisddavies / loan_approval_predictions

Loan Approval Predictions

This notebook is just playing around with a dataset provided at Kaggle.

About the Data

The data provides information such as income, loan amount, worth in assets, number of dependents, self-employed status, etc, along with labels for loan status. This data was used to make accurate predictions for which applicants would be accepted for loan repayments.

Data Cleaning and Preparation

The data was cleaned by performing the following:

• Scaling the continuous data
• Added an extra column for income category
• Encoded discrete values to One Hot Vectors
• Removed the loan_id column

Models Used

The following models were used on the dataset:

• Decision Tree
• Random Forest
• Logistic Regression
• Shallow Neural Network (approximately 30 nodes)

Findings

• The mutual information for just the cibil_score column was very high, and thus had very high correlation with the loan approvals
• When the cibil_score column was removed, the models performed poorly for predictions, either showing huge overfitting or underfitting
• Decision Tree performed the best, because obviously it could almost create a decision stump on just the cibil_score alone and predict very accurately