Machine Learning-Based Loan Default Prediction: Models, Insights, And Performance Evaluation In Peer-To-Peer Lending Platforms

Abstract

In financial institutions, mitigating capital loss is paramount, especially when considering the risks of extending loans. It's crucial to analyze potential dangers and thoroughly assess the likelihood of default to address these risks. Despite possessing extensive customer behavior data, financial institutions often need help accurately predict the loan default probabilities. Data mining, a rapidly advancing field in data analysis, offers promising solutions by extracting valuable insights from complex datasets. This research aims to develop and prototype a classification model based on deep learning algorithms, leveraging tools provided in the statistical tool Python. We preprocess the raw dataset to remove unimportant dimensions, detect outliers, remove them, input missing values, and normalize data to enhance prediction accuracy. Once we develop the model, we implement it to predict outcomes using a test dataset. Experimental findings confirm its accuracy in forecasting loan defaults.