The application of machine learning in bank credit rating prediction and risk assessment

Bank rating prediction can be important for financial environment in any country. With the complex dimension of datasets, it requires effective dimensionality reduction method and high accurate model. Among machine learning, recursive feature selection, lasso regression and principal component analysis (PCA) may have better ability in dimensionality decrease. And supervised learning includes random forest, SVM and artificial neural network attract great attention by high accuracy and powerful adaptation. Our study presents a combinational feature selection procedure by using the lasso and recursive feature elimination, in order to Figure out the key features among the bank credit rating prediction. Three important features are extracted and applied to the training of random forest, SVM and gradient boosted classification. Finally, SVM model gain the best accuracy of 86% in validated dataset. In addition, in order to evaluate the uncertainty provided by complex and intricate finical environment, we apply proportional hazards model (COX) and Kaplan-Meier to explain the effect of each variables. We measure the possible relationship between a company’s daily revenue and its credit ranking. It implies that the days of zero income and negative income may affect the company’s credit rating.