The “curse of dimensionality” issue caused by high-dimensional datasets not only imposes high memory and computational costs but also deteriorates the capability of learning methods. The main purpose of feature selection is to reduce the dimensionality of these datasets by discarding redundant and irrelevant features, which improves the performance of the learning algorithm. In this paper, a new feature selection algorithm, referred to as FSBACOM, was presented based on the binary ant system (BAS). The proposed method sought to improve feature selection by decreasing redundancy and achieved an optimum solution by increasing search space in a short time. For this purpose, the features were organized sequentially in a circular graph, where each feature was connected to the next one with two select/deselect edges. The proposed representation of the search space reduced computational time significantly, particularly on the high-dimensional datasets. Inspired from genetic algorithm and simulated annealing, a damped mutation strategy was introduced to avoid falling into local optima. In addition, a new idea was utilized to reduce the redundancy between selected features as far as possible. The performance of the proposed algorithm was compared to that of state-of-the-art feature selection algorithms using different classifiers on real-world datasets. The experimental results confirmed that FSBACOM significantly reduces computational time and achieves better performance than other feature selection methods.
