The Strength of Convolutional Neural Network in Financial Distress Prediction

Aim of study:

 In this study, by reviewing at the literature of financial distress prediction, a deep learning method has been developed that automates and shortens the earlier procedures of financial distress prediction. At last, the developed method is validated by comparing it to the earlier methods of this area.

The developed Convolutional Neural Network is used to extract knowledge within 14 related features to financial distress, through 3 consecutive years. In order to avoid the probable generalization error, the advantage of K-fold method to slice the entire 300 sample into sub-samples, was taken. Finally, the advantage of non-parametric Wilcoxon test was taken to verify the differences between the accuracy of the proposed method with the earlier ones.

Results of this study implies that the Convolutional Neural Network beats the other well-known methods in this area such as logistic regression and support vector machine with 95 percent confidence interval, in terms of prediction accuracy.