Performance Comparison of a Similarity-based Learner, a Genetic Classiier System, and a Hybrid Learning System

The Hybrid Learning System (HLS) inductively learns concepts and is a hybrid of a genetic classiier system and a similarity-based learner. Its basic structure is a genetic algorithm that applies a local search operator to population members. This study compares ID3, a standard genetic classifer system, and HLS for accuracy and concept conciseness on a variety of problem domains with and without noise. The results show that the genetic learners have superior accuracy and conciseness, except on randomly constructed artiicial concepts. Although the genetic learners perform similarly, the local search operator of HLS gives its genetic algorithm search faster convergence, allowing it to obtain similar or better results than an ordinary genetic search with less relative eeort.