A combination use of bagging and random subspace with memory mechanism for dynamic financial distress prediction

Abstract

On account of great significance of financial distress prediction for corporations, it is essential to construct an effective prediction model for managers and investors. Traditional financial distress prediction methods design static models using samples within a period of time, but the static models are insensitive to changes, such as concept drift in financial distress. This paper proposed a dynamic prediction framework called multi-layer perceptron ensemble method with memory (MPLE-FM). To improve the prediction performance of the model, two widely used ensemble method, bagging and random subspace, are combined to perturb both the instance space and feature space of the data to get diversity among classifiers. Based on data from the Chinese listed companies' real data from 2001 to 2014, the results showed that the proposed method has a better prediction performance over the other four dynamic prediction methods.