Research on Loan Default Prediction Based on Survival Analysis Method

Against the backdrop of slowing global economic growth and heightened market uncertainty, financial market participants’ ability and willingness to perform are facing severe challenges, and credit risks have increased significantly. In order to improve the accuracy of loan default discrimination and realize the effective life cycle management of default risk, this paper proposes a dynamic loan default credit risk assessment method, namely the CatBoost-SurvTRACE fusion model, by integrating machine learning algorithm and deep survival analysis method. Based on the theory of lifecycle management, this paper divides the life cycle of loans into three periods: early, middle, and late. Then, the CatBoost algorithm is used to analyze the importance of the features at each stage, and the core feature set is selected as the input of the SurvTRACE model, so as to analyze and predict the survival of the default risk. Experimental results show that the fusion model performs well in dealing with high-dimensional, complex, and non-equilibrium credit data, and its performance is better than other benchmark models in loan default discrimination and risk prediction tasks at different time points.