Binary DOA with Bell-Shaped Transfer Function

BDOA uses Bell-shaped transfer functions for feature selection, improving accuracy, fitness, convergence, and feature reduction.

This research introduces the Binary Dhole Optimization Algorithm (BDOA), a novel binary optimization method for feature selection problems inspired by the cooperative hunting behavior of dholes. Feature selection is an important task in machine learning because it can reduce computational cost, improve classification performance, and eliminate irrelevant and redundant features. However, selecting an optimal subset of features remains a challenging combinatorial optimization problem, especially for high-dimensional datasets. To address this issue, this study extends the original Dhole Optimization Algorithm into the binary domain through the use of novel Bell-shaped transfer functions. These transfer functions map continuous search dynamics into binary decisions while maintaining a balance between exploration and exploitation. Unlike traditional transfer functions, the proposed Bell-shaped transfer functions help reduce excessive bit flipping, preserve population diversity, and avoid premature convergence.The proposed BDOA was evaluated using multiple benchmark datasets and compared with several well-known binary optimization algorithms. The experimental results show that BDOA achieved competitive classification accuracy, lower fitness values, and fewer selected features across different datasets. In particular, BDOA achieved strong performance on datasets such as Ionosphere, Breast Cancer WD, Iris, Lung Cancer, and Human Activity Recognition. The results demonstrate that the proposed Bell-shaped transfer functions improve convergence behavior, computational efficiency, and solution quality. Quantitatively, BDOA improved the average classification accuracy from 0.8282 for the original DOA to 0.8718, representing an absolute improvement of 4.36 percentage points. In terms of mean fitness value, BDOA achieved the lowest average value of 0.0786 across all datasets, compared with 0.0896 for the closest competing method, representing an absolute reduction of 0.0110. Therefore, BDOA can be considered an effective and robust approach for binary optimization and feature selection problems. The source code of the proposed BDOA is publicly available at: https://github.com/bnyad95/Binary-Dhole-Optimization-Algorithm-BDOA