Machine Learning-Based Loan Approval Prediction and Risk Analysis

Loan approval depends on the information collected from an applicant. When some of that information is missing, a prediction model may give a different result. This study develops a simple machine learning application for loan approval prediction and examines how reduced applicant information affects the model. A public dataset containing 614 loan records is used. The data is cleaned, missing values are filled, categorical fields are encoded, and the input features are scaled for Logistic Regression. Logistic Regression is used as the main research model, while Random Forest is kept as a second model in the application. With complete information, Logistic Regression gives an accuracy of 86.18%. The accuracy becomes 85.37% at 5% missing information, 83.74% at 10% and 20%, and 79.67% at 30%. The overall fall from the baseline to the 30% setting is 6.51 percentage points. The work also includes feature-wise testing and decision stability as part of the planned analysis. The results show that the amount of applicant information supplied to the model can affect prediction performance.