Binary Horse herd optimization algorithm with crossover operators for feature selection.

This research article presents a binary version of the horse herd optimization algorithm (BHOA) for use in feature selection (FS). The binary version was arrived at by adapting the transfer function of the original algorithm and adding one-point, two-point, and uniform crossover operators. After testing the algorithm on multiple datasets, the researchers found that their binary version achieved highly competitive or superior results compared to leading methods in the field.

Main Objectives

• The primary goal of this study was to create a binary version of the Horse Herd Optimization Algorithm (HOA). This algorithm is based on the behavior of a pack of horses and its survival tactics.
• The researchers aimed to adjust the model in two significant ways: by using different transfer functions to turn the continuous domain into a binary one, and by adding crossover operators to ensure efficiency.
• They tested 15 versions of their proposed BHOA methodologies on 24 real-world datasets and compared the outcomes against 21 modern methods.
• Their final goal was to identify the best-performing version of their proposed algorithm, with results assessed on six metric measures.

Methodology

• The researchers mimicked the conduct of a horse herd in survival mode to create the binary algorithm. The algorithm iteratively optimizes solutions by simulating the social behavior of the horse herd.
• They used three transfer functions – S-shape, V-shape, and U-shape – to effectively transform the continuous domains into binary ones, yielding four alternatives for each function.
• Three crossover operators – one-point, two-point, and uniform – were incorporated into the binary algorithm to ensure its effectiveness for the feature selection (FS) domain.
• Their proposed methodology was then tested using 24 real-world datasets and evaluated using six metric measures: Accuracy, number of selected features, fitness values, sensitivity, specificity, and computational time.

Main Findings

• The proposed Binary Horse Herd Optimization Algorithm (BHOA) with S-shape transfer function and one-point crossover performed the best among 15 versions, delivering superior or highly competitive outcomes when benchmarked against 21 other methods.
• The researchers found that their proposed method was able to yield comparable and in some cases better-recorded results, confirming its effectiveness and efficiency for feature selection.
• As a result of its effectiveness and viability, the authors suggest that their proposed methodology could be applied to other areas of machine learning beyond feature selection.