Nature inspired feature selection meta-heuristics

摘要: Many strategies have been exploited for the task of feature selection, in an effort to identify more compact and better quality subsets. A number evaluation metrics developed recently that can judge a given subset as whole, rather than assessing qualities individual features. Effective techniques stochastic nature also emerged, allowing good solutions be discovered without resorting exhaustive search. This paper provides comprehensive review most recent methods selection originated from inspired meta-heuristics, where classic approaches such genetic algorithms ant colony optimisation are included comparison. reviewed methodologies significantly modified present, order systematically support generic subset-based evaluators higher dimensional problems. Such modifications carried out because original studies either work exclusively with certain (e.g., rough set-based methods), or limited specific problem domains. total ten different examined, their mechanisms flows summarised unified manner. The performance compared using high dimensional, real-valued benchmark data sets. selected subsets used build classification models, further validate efficacies.