Multi-Sub-Swarm PSO Classifier Design and Rule Extraction

Abstract

In this paper, a new classifier, based on multi-sub- swarm PSO algorithm, is proposed. Natural number coding is used in the classifier to avoid the updating inconvenience of binary encoding that has different properties dimensions. Classification is done by parallel search of multi-sub-swarm PSO algorithm. According to the characteristics of the coal mine gas emission concentration data, an extraction model is constructed of classification rules of coal gas emission concentration. The results showed that this classifier have high prediction accuracy rate, and the gas emission concentration rules, extracted from its rule space using this classifier, run efficiency significantly with less redundancy.