A SOM and GP tool for reducing the dimensionality of a financial distress prediction problem

Abstract

In order to build prediction models that can be applied to an extensive number of practical cases we need simple models which require the minimum amount of data. The Kohonen's self organizing map (SOM) are usually used to find unknown relationships between a set of variables that describe a problem, and to identify those with higher significance. In this work we have used genetic programming (GP) to produce models that can predict if a company is going to have book losses in the future. In addition, the analysis of the resulting GP trees provides information about the relevance of certain variables when solving the prediction model. This analysis in combination with the conclusions yielded using a SOM have allowed us to reduce significantly the number of variables used to solve the book losses prediction problem while improving the error rates obtained. © 2008 Springer-Verlag Berlin Heidelberg.