Evolutionary optimisation of classifiers and classifier ensembles for cost-sensitive pattern recognition

Pattern recognition problems occur in many fields and hence effective classification algorithms are the focus of much research. In various circumstances not classification accuracy but misclassification cost minimsation is the primary goal leading to the development of cost-sensitive classification algorithms. In this paper, we show how evolutionary algorithms, in particular genetic algorithms (GAs), can be employed optimise to cost-sensitive classifiers and classifier ensembles. In particular, we discuss how GAs can be employed to derive a compact set of fuzzy if-then rules with an embedded cost term, and how GAs are able to perform simultaneous classifier selection and fusion for ensemble classifiers.