Combining Feature Selection and Classification Using LASSO-Based MCO Classifier for Credit Risk Evaluation

Credit risk evaluation is a difficult task to predict default probabilities and deduce risk classification, and many classification methods and techniques have already been applied in predicting credit risk. In this paper, in view of the significant limitations of feature reduction and weak interpretability of the multi-criteria optimization classifier (MCOC), an improved LASSO-based MCOC (LASSO-MCOC) for simultaneous classification and feature selection is proposed and the corresponding algorithm is constructed. Based on the four real-world credit risk datasets, the LASSO-MCOC with linear and RBF kernels are tested and compared with the SMCOC proposed by Zhang et al. (2019) and six basic classification methods including logistic regression, multilayer perceptron, support vector machines, Naïve Bayes, k-nearest neighbors and random forest. The experimental and statistically comparative analysis results show that the LASSO-MCOC we proposed is more effective for credit risk assessment with better performance in accuracy, efficiency, and interpretability than that of other classifiers and can be extended to other real-world applications.