Whale Optimization Algorithm for High-dimensional Small-Instance Feature Selection

摘要

In this paper, two variants of the Whale Optimization Algorithm (WOA), called SWOA and VWOA, are introduced and used as search strategies in a wrapper feature selection model. Feature selection is a challenging task in machine learning process. It aims to minimize the size of a dataset by removing redundant and/or irrelevant features, with no information lose, to improve the efficiency of the learning algorithms. In this work, two transfer functions (i.e., sigmoid and tanh) that belong to two different families (S-shaped and V-shaped) are used to convert the continuous version of the WOA to binary. The proposed approaches have been tested on 9 different high dimensional medical datasets, with a low number of samples and multiple classes. The results revealed a superior performance for the VWOA over the SWOA and other approaches used for the comparison purposes.