ABCN/BFS: Modified Artificial Bee Colony Feature Selection with Genetic Operators
Year: 2026
Type: article
Abstract: Feature selection is a fundamental preprocessing strategy in contemporary data-intensive applications. By identifying and removing redundant or weakly informative features, the model's accuracy improves, its generalization capability is enhanced, and computational complexity is reduced. Obtaining features that accurately reflect the characteristics of the datasets and eliminating features that cause errors is crucial for achieving accurate results. The detection of eligible features through these systems can be utilized in various modern applications, including game development, product development, and advertising. While the Artificial Bee Colony algorithm is a well-established metaheuristic, its application to discrete feature selection often suffers from premature convergence and a lack of diversity in high-dimensional search spaces A persistent gap in the literature is the need for a low-cost, binary-native modification that maintains the adaptive search concept of the original ABC without the excessive complexity seen in recent ensemble-based or multi-stage hybrid models. This research introduces the ABCN/BFS framework, consisting of two distinct approaches: ABCNFS, which utilizes a numerical modification rate (MR = 0.5) to control parameter oscillation, and ABCBFS, which implements a binary search strategy based on 85% two-point crossover and 15% single-point mutation. These models were evaluated against 24 UCI benchmark datasets using a weighted objective function and 10-fold cross-validation. The proposed approaches were compared with studies using similar representations and techniques in the literature, using the same sample datasets, and their performance examined. The results yielded successful results on benchmark datasets, and their performance is discussed.
Source: Bilişim Teknolojileri Dergisi
Authors: Turgay Batbat, Ayşegül Güven
Institution: Erciyes University
Language: Turkish
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7159777243