Field Learning


Honghua Dai
Department of Computer Science, Monash University, Clayton 3168, Australia.
dai@bruce.cs.monash.edu.au


Abstract

Low prediction accuracy (LPA) problem can be caused by several factors. Overfitting low quality data and being misled by them seem to be the significant ones. Traditional inductive learning algorithms derive rules by working on point values of each attribute. This approach could easily be misled by low quality data and the derived rules could overfit them and cause LPA problem. This paper presents an alternative way which derives rules by working on the fields of attributes with respect to classes, rather than on on individual point values of attributes. The experimental results results show that field learning achieved a higher prediction accuracy rate and a higher prediction ability rate on new unseen test cases which is particularly true when the learning is performed on large low low quality data.
Conference Home Page