Predict Default Risk of P2P Loans
In this 2-year project, we delivered machine learning solutions to a start-up company that provided P2P loan liquidation services.
We built XGBoost models in R to predict the default risk of lending club loans at issuance. Model training and cross validation were performed using parallel computing on AWS cloud instances. We also built API services using the R plumber package to enable communications between the database, models, and dashboard.
We analyzed the life cycle of primary loans and developed a formula for calculating the fair price of seasoned loans to sell on the secondary market. We also produced technical white papers and marketing content.
The primary notes purchased by our models had 7 ~ 23% fewer defaults or late payments than auto-investing on lending club. And our pricing formula allowed quick and profitable sales of seasoned notes on the secondary market.