GA-based optimisation of fuzzy rule bases for pattern classification

Many decision making problems can be formulated as pattern classification problems. Therefore, high performing classification algorithms are highly sought after. Rule based pattern classification algorithms have an advantage that they do not appear to the user just as a “black box” but may provide additional insight based on the generated rules. In this paper, we focus on fuzzy rule based approaches which employ concepts from fuzzy logic theory to encode input patterns in a non-binary way. Starting with a basic fuzzy classifier we show that, through a simple modification, it can be turned into a cost sensitive classification method, and that classification performance can be improved through an error correction learning approach. Importantly, since rule-based classifiers are prone to rule explosion, we then show that a compact yet powerful rule base can be generated through an optimisation approach based on genetic algorithms.