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

Area of Science:

• Machine Learning
• Optimization Algorithms
• Computational Intelligence

Background:

• Feature Selection (FS) is crucial for improving model performance and reducing complexity.
• Existing optimization algorithms may not be optimally suited for binary FS problems.
• The Horse Herd Optimization Algorithm (HOA) is a nature-inspired metaheuristic.

Purpose of the Study:

• To develop and evaluate a binary version of the Horse Herd Optimization Algorithm (BHOA) for feature selection.
• To investigate the impact of different transfer functions and crossover operators on BHOA's performance.
• To compare the proposed BHOA variants against established state-of-the-art feature selection methods.

Main Methods:

• A binary version of the Horse Herd Optimization Algorithm (BHOA) was developed.
• Three transfer functions (S-shape, V-shape, U-shape) with four configurations each were employed.
• Three crossover operators (one-point, two-point, uniform) were integrated.
• Fifteen BHOA versions were tested on 24 real-world FS datasets.
• Performance was evaluated using accuracy, feature count, fitness, sensitivity, specificity, and computational time.

Main Results:

• The BHOA with S-shape transfer function and one-point crossover emerged as the best-performing version.
• The proposed BHOA method achieved highly competitive results, with some being the best recorded.
• Comparative analysis showed superior or comparable performance against 21 state-of-the-art methods.
• The optimized BHOA effectively balances feature reduction with classification accuracy.

Conclusions:

• The developed Binary Horse Herd Optimization Algorithm (BHOA) is a viable and effective method for feature selection.
• The combination of S-shape transfer function and one-point crossover yields optimal results for BHOA in FS tasks.
• BHOA shows significant potential for application in various machine learning domains beyond feature selection.