Vehicle Loan Default Prediction Dashboard

The Challenge

Predicting loan default from a highly imbalanced dataset (22% default rate) required careful model selection and an accessible way for non-technical users to evaluate results.

The Approach

• Benchmarked six ML classification models (including XGBoost) on ~230K imbalanced records.
• Deployed an interactive Streamlit dashboard supporting CSV upload, confusion matrices, and real-time evaluation metrics.

Results

• Achieved 0.629 AUC on a highly imbalanced classification problem.
• Shipped a live, publicly accessible dashboard for real-time model evaluation.