Loan Risk Prediction Analysis
A predictive analytics pipeline that scores loan applicants by default risk using a mix of exploratory data analysis, feature engineering, and classical ML models. Built in Python to give lending teams a defensible, explainable risk signal before underwriting.
The problem
Loan officers needed a faster, more consistent way to estimate default risk for new applicants. The existing process leaned on spreadsheets and tribal heuristics, which made underwriting slow and made it hard to explain decisions to auditors.
The approach
I pulled together applicant features across experience, education, industry, and location, then ran a full EDA pass in pandas and seaborn to understand distributions, outliers, and feature interactions. From there I trained and compared regression and tree-based models, settling on the configuration that balanced precision-recall tradeoffs for the lender's risk tolerance.
Stack
Python, pandas, NumPy, seaborn, Matplotlib, scikit-learn. Workflow ran inside Jupyter for iteration and was packaged as a reusable scoring module for downstream integration.
Outcomes
Cut manual triage time on new applications dramatically and gave the team a single, model-backed risk score they could defend with feature-importance reports. The pipeline also became the template for follow-on credit-modeling projects.