Genetic Search for Accurate Feature Sets

Abstract

This abstract describes a feature selection system, INDiGENT. INDiGENT has been to designed to enhance knowledge based neural networks by genetically searching the set of input features for an optimum subset. This search is designed to enable INDiGENT to make more accurate classifications. Signifigantly, INDiGENT has shown that it can obtain similar increases in accuracy as more complicated theory revision systems. Expert systems have proven themselves effective decision makers for many types of problems. However, the accuracy of such systems is highly dependent upon the accuracy of the human expert’s domain theory. To escape this dependency, many machine learning systems have been developed to automatically refine and correct an expert’s domain theory. Classification systems rely heavily upon having the best (i.e., most relevant) set of input data. Many domains havie a vary input features. Therefore, choosing the appropriate set of inputs is critical to the performance of any expert system, and specifically in this case, to the creation of accurate neural networks. There are a number of systems which use neural network encodings to refine a variety of rule bases including finite state automata (Maclin & Shavlik 1993), certainty-factor rule bases (Mahoney 1996) and firstorder horn clause logic (Towell, Shavlik, & Noordewier 1990).