Binary Fish School Search applied to Feature Selection

The aim of the present paper is to develop efficient feature selection approaches. A novel wrapper methodology for feature selection is formulated based on the Fish School Search (FSS) optimization algorithm, intended to cope with premature convergence. In order to use this population based optimization algorithm in feature selection problems, the use of binary encoding for the internal mechanisms of the fish school search is proposed, emerging the binary fish school search (BFSS). The proposed algorithm, as well as other state of the art feature selection methods such as Sequential Forward Selection (SFS) and Binary Particle Swarm Optimization (BPSO), were combined with fuzzy modelling in a wrapper approach and tested over two databases, a benchmark and an ICU (intensive care unit) database. The purpose of using this last database was to predict the readmission of ICU patients 24 to 72 hours after being discharged. Several statistical measures were considered to characterise the patient stay, including the Shannon entropy and the weighted mean. The results obtained by comparing the performance measures and the number of features selected of the used algorithms, show promising results for the novel algorithm BFSS.