Bankruptcy Prediction Of Non-life Insurance Enterprises Based On LightGBM Model With Feature Engineering Application

COVID-19 is spreading around the world,and a series of emergencies are changing the competitive landscape of the insurance industry.Non-life insurance companies are facing multiple transformation pressures and even triggering bankruptcy crises.Non-life insurance enterprises guarantee the property safety of the majority of insured subjects,but also closely related to the stability of the national economy,and the traditional method of bankruptcy prediction is limited because of too many assumptions,so it is of great significance to use the model to predict the bankruptcy of non-life insurance enterprises.Based on feature engineering,descriptive statistical analysis,data cleaning,feature construction,feature selection,feature extraction and other processing are carried out on the data of high-dimensional non-life insurance enterprises.From the perspective of feature engineering application,this paper sorts out and summarizes relevant research literature on the bankruptcy theory and bankruptcy prediction methods of non-life insurance enterprises.The global non-life insurance enterprise bankruptcy prediction Light GBM model is constructed by the integration algorithm of gradient elevation.Finally,the feature engineering model and the untreated control model are compared and evaluated,and the SHAP prediction results of the feature selection group model are interpreted by the method.Through model comparison and interpretability analysis,the following conclusions are drawn:First,the prediction accuracy of each group of samples treated by feature engineering is significantly improved compared with the control group,indicating that each step of feature engineering has a positive impact on the prediction accuracy of the Light GBM model.Secondly,the median and value of the evaluation indexF₁ and AUC for classification prediction by feature construction group Light GBM model improved more.The AUC value,recall rate and accuracy rate of feature selection group were increased by more than 10%.The accuracy rate and recall rate of feature extraction group had the most obvious improvement effect in feature construction,feature selection and feature extraction,and the improvement effect was about 15%.Finally,the SHAP explanatory method was used to reverse interpret the model,and the characteristic variable"total debt and earnings"contributed the most to the classification prediction results of the feature selection group model.