Credit Risk Analysis using LightGBM and a comparative study of popular algorithms

Abstract

Credit Risk analysis and mitigation have been an area of concern since the 07–08 Financial Crisis. One of the main reasons for the collapse was the high default rates of low-income security loans. Calculating credit scores can be a complicated process for people with thin credit histories or non-existent credit histories. Banks may refuse to give loans if the scores don't satisfy their requirements. Lack of a credit score is considered as an indicator for potential default and hence banks avoid sanctioning loans for people who come under this category. However, banks still offer loans if people are willing to offer securities. Credit Scoring can be done by using state-of-the-art Machine Learning models. Machine Learning and Data Science are becoming increasingly crucial in the fin-tech world. Popular machine learning algorithms such as Random Forest and Linear Support Vector Machines are being used currently. We're looking to explore further into credit risk analysis with LightGBM as our algorithm of choice. It is an open source framework developed by Microsoft in 2017. It is an ensemble model which has several advantages such as better prediction and higher stability. Predictions aggregated from multiple models tend to be less noisy than a single model, this is one of the main reasons why an ensemble model such as LightGBM can perform better than Logistic Regression and other algorithms like SVMs for this use case.