Collection Intensity Scoring and Channel Recommendation

Project Goal: The goal of this project is to develop and implement a comprehensive solution to optimize the collection process for peer-to-peer (P2P) lending. This includes creating a collection intensity scoring model to evaluate and prioritize collection efforts based on customer repayment risk. It also includes a predictive model to recommend the most effective communication channels with their number of customer interactions.

Task: Performed Extract, Transform, Load for data integration using PostgreSQL, developed a Random Forest classifier to categorize collection intensity and a K-Nearest Neighbors Regressor for predicting interaction counts in recommended channels. Implemented model optimization using Grid Search and Bayesian Optimization in Python, and conducted statistical tests using R. Results showed Bayesian Optimization’s superior performance in both models, achieving 98.34% accuracy and an MAE of 0.24530.