Application of Artificial Intelligence Algorithm of Linear and Non-linear Relevance Vector Machine in Predicting the Bankruptcy

Abstract

Abstract The purpose of this research is localization of a method for analyzing and predicting the Bankruptcy of companies at three levels (financial health, Bankruptcy and bankrupt). In the first step, using the Relief-F multi-class AI algorithm among 54 initial independent variables, and using information from 1488 companies-years during the period 2011-2016, the financial risk variables, working capital ratio, long-run debt ratio, asset flow ratio, the economic value added ratio, the ratio of non-executive managers, the ratio of current debts to equity, the ratio of debt to equity, corporate size, earning management were selected as important variables in the prediction of the three-level Bankruptcy situation, respectively. Using relevance vector machine algorithm, the Bankruptcy situation of companies is predicted in the coming year and next two years using MATLAB 2017 software. The results of the research indicate that in general, the predictive power of the relevance algorithm in nonlinear mode is much higher than in linear mode, so that in the nonlinear mode, using the relevance vector machine algorithm, we can determine the company's Bankruptcy with accuracy of more than 93% for the coming year and more than 86% for the next two years.