An Evolutionary Threshold Logistic Regression Model for Business Failures Forecast

A more precise business failure forecast helps to provide important judgment principles to decision-makers. The logistic regression (LR) model is a common evaluation model employed in prediction of business failures. Fewer researches focus on the threshold selection, which will influence the final predicting performance. Besides, it is very challenging to discover critical information from these statements, although financial statements reflect a certain degree of firm's business activities. However, selecting predictor variables depends on their linear searching characteristic, if applying traditional statistical methods, such as stepwise and etc. Genetic algorithm (GA) has been shown to be able to select the variable set with population diversity and to perform efficient search in large space. In this paper, we combine both LR and GA to form a new efficient business failure forecasting mechanism which maximizes the forecasting accuracy with minimum critical predictor variable and optimal threshold. From our experimental results, this approach effectively helps improving the forecasting accuracy of traditional LR.