Predicting Loan Default Likelihood Using Machine Learning
MetadataShow full item record
Traditional models that calculate credit worthiness are heavily dependent upon individual credit score or credit report information. This type of model makes it extremely difficult for individuals with little to no credit to obtain financial services, and if they are able to obtain these services, it is likely with very high interest rates and fees. To expand services to these individuals with more favorable rates while keeping the risk at acceptable levels, financial companies have been trying to figure out ways to implement non-traditional data into the credit risk process. Through the use of machine learning, non-traditional data is used to determine an individual’s credit worthiness with an accuracy that is acceptable to financial businesses. This research project performed data processing and feature engineering on data provided by Home Credit Group. This data was then used to train a Light GBM machine learning model and was able to predict the likelihood of default with an AUC of 0.7759.