vikrambansall / credit-risk-model-and-analysis

Credit Risk Assessment Using Advanced Machine Learning Techniques

Introduction: In the realm of financial services, predicting and mitigating credit risk is paramount to safeguarding the interests of institutions and clients alike. Leveraging the power of machine learning, we delve into the intricate world of credit risk analysis to identify potential defaulters and fortify financial stability.

About the Data: Drawing insights from a rich dataset provided by Nubank, a leading digital bank in Latin America renowned for its data-driven approach, we embark on a journey to analyze 43 key features across 45,000 client profiles. Our target variable, 'target_default', presents a dichotomy of True/False values representing the risk of default.

Machine Learning Models: Exploring the efficacy of cutting-edge gradient boosting algorithms, namely XGBoost, LightGBM, and CatBoost, we aim to discern the most adept model for predicting credit risk. These algorithms serve as invaluable tools in our pursuit of precision and recall, two pivotal metrics in the realm of credit risk analysis.

Results: Our endeavor to optimize model performance revolves around striking a delicate balance between false positives and false negatives. While XGBoost showcases commendable recall at 81%, it grapples with a relatively high false positive rate of 56%. Conversely, LightGBM and CatBoost exhibit superior false positive rates at 38% and 33% respectively, albeit at the expense of higher false negatives.