The Research On SGL-SVM Method And Its Application

The arrival of big data times makes the machine learning methods the new darling of the times.Support vector machine(SVM),as a high efficiency classifier,is applied to all walks of life.The traditional support vector machine is like a black box,and it cannot select the variables.Therefore,the author combines the support vector machine with the variable selection method based on the penalty function to achieve the purpose of the variable selection in the support vector machine.The current method can only achieve the purpose of the group variable selection in the support vector machine.Based on the previous research,the bi-level variable selection method,Sparse Group Lasso method,is combined with SVM in this paper,and the SGL-SVM method is proposed,which can select group variables between groups and select the variables in the groups.There is a better performance in the data existing sparseness among groups and sparseness in the groups.The optimization problem is solved by the coordinate gradient descent method and the block coordinate descent method.The AUC value is used as the model evaluation criterion.The optimal parameters are selected by the five-fold cross validation.In the three simulation experiments,we compared the SGL-S VM with the GL-S VM,L1-SVM and CSVM methods in the prediction effect and the variable selection effect.We found that the SGL-S VM performed well in the data with the group structure,especially in high dimension.The variable selection effect of SGL-SVM is improved while improving the prediction accuracy.The performance of SGL-S VM is stable in the different parameters and different group structure.In this paper,the SGL-SVM method is applied to the manufacturing industry financial distress forecast of listed companies.The ST listed companies of manufacturing industry in the A shares of 2013 to 2016 due to the financial distress are treated as the positive examples,and the remaining A-share manufacturing industry listed companies are negative cases.We balance the data by double sampling method with over-sampling and under-sampling,and extrapolated the 30%of the original data as the prediction.The accuracy of the positive examples of SGL-SVM method is much higher than that of the other methods.The comprehensive performance is also better.In conclusion,the ability to grow,profitability and revenue quality are major aspects of the company's financial situation.Cash flow and capital structure are also important indicators,and the importance of ability to report solvency and operational capacity are relatively limited.