Performance Study of Ant Colony Optimization for Feature Selection in EEG Classification.

Feature Selection (FS) is a contemporary challenge for the scientific community since new methods are being discovered and new forms of algorithmic design are required. In this sense, classic bio-inspired swarm intelligence algorithms can be explored to get a new suitable feature selection application where simple agents interact locally with each other while searching a global solution. Therefore, this work proposes an innovative utilization of the ant colony optimization algorithm for FS (ACO-FS) with the objective of reducing the high number of features present in Electroencephalogram (EEG) signals. Specifically, a base ant colony optimization algorithm and two variants have been implemented and evaluated in terms of execution time, energy consumption, and classification rates to point out the strength and weakness of each variant. The preliminary results demonstrate that the proposed method provides a classification rate close to 86% when the whole dataset is reduced from 3,600 features to only 22.