Semi-Supervised Learning Classifier System Based on Bayes

Abstract

The high interpretability and the extraordinary evolvability of learning classifier system make it the optimal choice to build an adaptive intelligent system, and UCS is one of its branches, which is especially designed for the supervised learning tasks. However usually there is a huge amount of unlabeled data that are helpful for the increasing of its accuracy. Hence we use the EM algorithm in Semi-supervised learning as a reference, and proposed a Semi-Supervised Classifier system (SUCS) based on Bayes inference. The experiments we did using the UCI dataset showed that SUCS performed a much better accuracy than UCS by use of only a small number of labeled data and a large number of unlabeled data.