Revolutionizing Credit Risk: A Deep Dive into Gradient-Boosting Techniques in AI-Driven Finance

Revolutionizing Credit Risk: A Deep Dive into Gradient-Boosting Techniques in AI-Driven Finance

연구 분야: Artificial Intelligence

학회: 2024 International Conference on Information Networking (ICOIN)

Assessing credit risk is essential for financial institutions to uphold responsible lending practices. In this study, we conduct an in-depth analysis of three state-of-the-art gradient-boosting algorithms—XGBoost, LightGBM, and CatBoost—for their applicability in credit risk assessment. Utilizing a complex 50 GB dataset with 2.3 million records and 190 features shared by the second-largest credit card issuer globally - American Express, we investigate various factors that influence prediction performance. Our research highlights that the size of data chunks plays a significant role in the algorithms’ performance, particularly noting that CatBoost performs exceptionally well with larger data segments. The study also emphasis the importance of effectively managing missing data, which has a marked impact on the capabilities of XGBoost and LightGBM. We also examine hyperparameter tuning to identify unique learning characteristics for each algorithm. In conclusion, our findings reinforce financial institutions with advanced analytical tools, enhancing their ability to make informed credit risk assessments.