A SMOTE-Lasso-Logistic Credit Scoring Model
Abstract
Credit scoring to classify good – bad borrowers is one of the important tasks of risk management at banks and credit bureaus. A reliable credit scoring model must correctly discover the bad class. This does not usually succeed if the difference of the number of good and bad borrowers is large. Besides, credit scoring model should point out the significant characteristics of borrowers to predict the probability of default. The paper proposes a credit scoring model called SMOTE-Lasso-Logistic. Applying the combination of the resampling technique SMOTE and Lasso method on Logistic regression, SMOTE-Lasso-Logistic model can solve these issues and have higher classification performance than traditional approaches such as Logistic regression and Decision tree model.