Designing a System for Bankruptcy Estimation base upon Neural Networks

Abstract

This paper deals with the problem to design a system for bankruptcy estimation based upon neural networks. At first, the design of multilayer neural networks having inputs of numerical financial ratios is discussed. It is preferable to control the optimization of variables in neural netowrks automatically and also it must be considered the performance of neural netowrks depends how we select the financial data, such as kinds of financial ratios and categories of business. Therefore, the system proposed in the paper utilizes subsystems which take specified action if a set of given conditions are satisfied (production rules). In the next step, a kind of reasoning is utilized to improve the output of multilayer neural network, on the basis of categorical data of companies. More precisely we select a group of companies for which the neural netowrk often give incorrect decision, and then we apply reasonig subsystem using production rules so as to increase correct decision. To compare the performance of the system treated here, the same examples for bankruptcy estimation are analyzed by the system and by the software package for multivariate analysis. As a result, it is seen the rate of correct estimation of the system is improved better than conventional multivariate analysis.