Efficient feature selection using a self-adjusting harmony search algorithm

Many strategies have been exploited for the task of feature selection, in an effort to identify more compact and better quality subsets. The use of an evaluation metric have been developed recently that can judge the quality of a given subset as a whole, rather than a combination of individual features. Powerful nature-inspired stochastic search techniques have also emerged, allowing multiple good quality features to be discovered without resorting to exhaustive search. Harmony search in particular, is a recently developed technique that mimics musicians' experience, which has been successfully applied to solving feature selection problems. This paper proposes three improvements to the harmony search algorithm that are designed to further enhance its feature selection performance. The resultant technique is more efficient, capable of automatically adjusting the internal components of the algorithm. Systematic experimental evaluation using high dimensional, real-valued data sets has been carried out to verify the benefits of the presented work.