XGBoost-based Survival Analysis in Business Risk Prediction

Business risk of enterprises have occurred frequently in economy area recently. Although many companies and scholars have built enterprise early warning system by binary classification or multi-class research, traditional machine learning models have poor time explanation, as they excessively pursue the prediction performance with complex machine learning approaches or deep learning models, may lead to some economic paradoxes of important risk factors, and deviates from the original intention of risk prediction. Therefore, we establish a novel non-linear survival analysis method, which not only provides a qualitative analysis of the key factors in business data, but also improves the prediction performance of XGBoost-based models. Impressive experiments have been conducted on the CSMAR database, results show the outperformance of our method compared to other approaches.