Finding Compact Classification Rules With Parsimonious Gene Expression Programming

Gene expression programming (GEP) is a new evolutionary computation technique. Its fixed-length representation spares it from the problem of bloat, which is a serious problem of GP. Recent work has shown that GEP can discover accurate and understandable classification rules in the form of mathematical expressions. But some of the rules are still too large than necessary because of the existence of useless parts in genes. Basic GEP classifiers don´t have the ability to remove the useless parts from the classification rules. This paper presents a new fitness function incorporating a parsimonious term to evolve more compact rules with parsimonious gene expression programming (PGEP). Results show that PGEP can find more parsimonious rules without losing accuracy