Integrated Algorithm-based Credit Risk Assessment and Credit Decision Guidance

Abstract

This paper investigates helping banks to assess the strength and creditworthiness of firms based on limited data on MSMEs, which leads to credit risk assessment and lending decisions. In the data of enterprise invoice with complete information, four indicators of annual profit, annual profit rate, annual profit growth rate and reputation rating were selected as independent variables (the last one symbolizes enterprise reputation, the other three symbolize enterprise strength), and whether to default or not as the dichotomous dependent variable, six single classifiers such as LDA, KNeighborsClassifier, NB, SVM, LR and MLP classifiers were trained with one The integrated classifier was finally selected through performance evaluation. For the data with missing reputation ratings, we use three indicators such as enterprise capital turnover, order completion rate, and effective invoicing rate from the data with complete information as features and reputation ratings as labels, and fit a classification prediction model by fisher linear discriminant, and then make predictions for the data with missing information. After excluding firms with a credit rating of D and a high risk factor, we allocated the loan amount to the total amount of the three types of firms according to their size. After fitting the equation for the relationship between APR and customer churn rate, a multi-objective nonlinear programming model was developed to solve the credit strategy with minimum customer churn rate and maximum profit as the objectives and loan amount and APR as the decision variables.