Sparse Maximum Margin Logistic Regression for Credit Scoring

The objective of credit scoring model is to categorize the applicants as either accepted or rejected debtors prior to granting credit. A modified logistic loss function is proposed which can approximate hinge loss and therefore the resulting model, maximum margin logistic regression (MMLR), has the classification capability of support vector machine (SVM) with low computational cost. Finally, to classify credit applicants, an efficient algorithm is also described for MMLR based on e.-boosting which can provide sparse estimation of coefficients for better stability and interpretability.