Research On Solvency Early Warning Of Property Insurance Company Based On Machine Learning

This study uses machine learning method to build early warning model for solvency of property insurance company,and identifies the best performance as the final early warning model from several selected machine learning methods.Although the solvency management of insurance companies is more standardized in general under the "second generation of solvency" regulatory rules,in recent years,some insurance companies’ solvency suddenly appears problems,which is extremely unfavorable for companies,policyholders and even the society.Establishing the solvency early warning model of property insurance company can not only make the company predict the solvency problem in advance and make improvement,but also improve the internal solvency management ability of insurance company.Based on the background of solvency and early-warning methods,according to the characteristics of early-warning model and the results of literature review,this paper constructs the early-warning indicators of solvency,and analyzes the scientificity and relevance of the selected indicators by using the gray correlation analysis method.In this process,due to the change of solvency requirements and the calculation of adequacy ratio in the "second generation of solvency" rules,the solvency adequacy ratio and the comprehensive risk level are re determined.Finally,select the data of 30 property insurance companies from 2010 to 2018,use the four selected machine learning classification methods to carry out model training,and use the trained model to predict the test set samples divided in advance.In this paper,the early warning results of four methods are analyzed and compared,and support vector machine is selected as the best method to build the solvency model.The empirical results show that the early warning index system established in this paper is more reasonable.Four machine learning methods predict test samples through training samples,but the prediction results are quite different.Among them,support vector machine model has the best performance,random forest and decision tree do not show good prediction ability.The prediction results of SVM model show that the training speed is fast and the prediction accuracy is high,especially the recognition rate of companies with solvency problems is much higher than other methods...