关键词: SVM, 双层变量选择, SGL, 财务困境预测

Abstract: For classification, we propose sparse group lasso SVM (SGL-SVM) approach to conduct bi-level variable selection by adding SGL penalty to SVM loss function. Due to the complexity of optimization of SGL-SVM objection function，we propose a bi-level coordinate descent algorithm. Simulation results suggest that SGL-SVM is better than other methods both on the performance of prediction and variable selection. For data with covariate group structure, bi-level selection not only can improve the performance of variable selection but also can improve the performance of prediction. Finally, we apply the proposed SGL-SVM to forecast manufacturing corporate financial distress in China.

Key words: SVM, Bi-level Variable Selection, Sparse Group Lasso, Forecasting Corporate Financial Distress