A comprehensive modeling method of continuous and discrete variables for personal credit forecasting

With the rapid development of the Internet economy, the range of personal credit has expanded and become an important research issue in the credit area. Existing data-driven approaches to personal credit forecasting do not consider different variable types in an integrated manner, leading to inaccurate results. This paper investigates the application of the decision tree-logistic regression hybrid model to the mixed data type problem. Our model first adopted the decision tree to process only the discrete variables and then output their corresponding probability. Next, we input the probability and other continuous variables into the logistic regression model. To test the reliability of our proposed model, we tested it on a personal credit default dataset. We have obtained a 79.9% accuracy in our proposed model, which is higher than using either the decision tree model or the logistic regression model alone. Compared to using the decision tree or the logistic regression model alone, our proposed model successfully coordinates the discrete and continuous variables and avoids information loss. Banks and financial institutions can use this model to assess the default risk of their loan customers and improve the quality of their loans.