Application Of A New Model Based On Boosting Algorithm In Bank Credit Scoring

In the era of Big Data, using credit scoring model to accurately judge the applicant’s credit risk based on big data provided by banks is the trend of future development. In the actual credit scoring database, there is a fact that applicants with "good" credit are usually more than those with "bad" credit, leading to an unbalanced data set. When training on those sets by machine learning, those classifiers always hardly recognize the "bad" sample. In fact, applicants wrongly given a "good" credit will bring the bank a huge business harm. Therefore, to improve the recognition of "bad" class is crucial.The paper proposes a new model HSBoost approach based on hybrid-sampling technology and Boosting approach. In phase one, it balances the training set with above hybrid-sampling approach. In phase two, it takes advantage of Boosting algorithm to improve the recognition ability of classifier for "bad" class.This paper uses credit scoring datasets of UCI databases to do HSBoost approach’s empirical analysis with SVM, BP and DT. Compared to RUSBoost, SMOTEBoost, hybrid-sampling, under-sampling and over-sampling, it is proved the feasibility and effectiveness of this algorithm. In addition, nonparametric Wilcoxon signed rank test is held to confirm that HSBoost is superior to SMOTEBoost.