A Novel Nonlinear Integrated Forecasting Model of Logistic Regression and Support Vector Machine for Business Failure Prediction with All Sample Sizes

Abstract

Abstract The aim of this work was to improve the forecasting performance of business failure prediction with all sample sizes by constructing a novel nonlinear integrated forecasting model (ANIFM) of individual linear forecasting models and individual nonlinear forecasting models. First, a new variable set including internal variables and external variables was proposed. Using scatter diagrams, all variables were placed in either the linear group or the nonlinear group. We considered logistic regression (LR) as the individual linear forecasting method to deal with each linear variable, the support vector machine (SVM) as the individual nonlinear forecasting method to deal with each nonlinear variable, and the residual SVM as the integration method to integrate the forecasts of LRs and SVMs. The proposed procedure was applied to real datasets from China. For performance comparison, single LR, SVM methods, integration forecasting models based on equal weights and on neural networks, and one based on rough set and Dempster-Shafer evidence theory (D-S theory) were also included in the empirical experiment as benchmarks. The experimental results demonstrate the superior forecasting performance of the proposed ANIFM in terms of forecasting accuracy and forecasting stability, especially with small sample sizes.