Generating AVTs Using GA for Learning Decision Tree Classifiers with Missing Data

Abstract

Attribute value taxonomies (AVTs) have been used to perform AVT-guided decision tree learning on partially or totally missing data. In many cases, user-supplied AVTs are used. We propose an approach to automatically generate an AVT for a given dataset using a genetic algorithm. Experiments on real world datasets demonstrate the feasibility of our approach, generating AVTs which yield comparable performance (in terms of classification accuracy) to that with user supplied AVTs. © Springer-Verlag 2004.