XGBoost Optimized by Adaptive Tree Parzen Estimators for Credit Risk Analysis

Credit Risk analysis of customers protects and ensures Banking institutions' interests and have been a core part of any financial system. Traditional Banks have been using decades old legacy systems for all of their processes and these systems are no match for current Machine Learning and Deep Learning algorithms. XGBoost is a tree-based distributed gradient boosting framework famous for its efficiency and flexibility. The performance of XGBoost largely depends on its hyperparameters which can be optimized. Methods like GridSearch and RandomSearch are largely used for hyperparameter optimization but are inefficient in nature. This paper proposes a XGBoost based algorithm optimized by Adaptive Tree Parzen Estimators (ATPE) for Credit Risk analysis. The hyperparameters of the proposed algorithm, when optimized with ATPE give better results compared to the default model and thus it can be used as a replacement for the legacy systems used in Banks for Credit Risk analysis.