Leveraging Machine Learning Techniques For Developing Robust Credit Scores For Peer-To-Peer Lending Platforms

Abstract

Peer-to-Peer lending platforms have become a widely accepted alternative to conventional banking, providing borrowers and lenders with a direct method of financial transaction. The study commences with an overview of P2P lending platforms, delving into the progression of credit scoring and providing a synopsis of the P2P lending market. A thorough examination of the existing literature establishes the foundation for our inquiry. We describe our data pre-processing procedures, which involve manipulating independent variables and filling in missing values, using a comprehensive dataset obtained from a prominent P2P lending platform. Subsequently, we employ automation to compute discrete and continuous variables to pre-process the data for analysis. The study progresses by splitting data into train and test datasets to calculate the probability of the default (PD) model. The logistic regression machine learning algorithm creates and verifies the credit score model. The model estimation process is subsequently validated using test data, where performance metrics such as Kolmogorov-Smirnov and Gini statistics are employed to estimate predictive accuracy and discriminatory power.

This study substantially improves credit scoring systems in P2P lending platforms by utilizing machine learning techniques. Our study strengthens lenders' risk assessment capacities, providing them with more powerful tools that promote trust and efficiency in lending. The results highlight the significant impact that machine learning may have on credit rating, enabling the sustainable expansion of the P2P lending sector.