An Experimental Design Approach to Analyse the Performance of Island-Based Parallel Artificial Bee Colony Algorithm

The Artificial Bee Colony (ABC) is a novel nature-inspired metaheuristic optimization algorithm that mimics the behavior of honey bees searching for food sources. The main drawback of ABC, similar to the most of metaheuristics, is the premature convergence (i.e., the earlier stuck into local optima). Recently, the structured population approach, in which the individuals are distributed into multiple sub-populations (called islands), has been widely exploited to maintain the required diversity during the search process and thus reducing the prematurity problem. In this paper, the island model, which is a common structured population approach, is incorporated with the ABC to introduce a parallel variant called (iABC). Besides, an experimental design approach is proposed to analyze the sensitivity of iABC to the parameters of the island model as well as the main specific parameters. The linear regression model and the Analysis of variance (ANOVA) are utilized to estimate the effect of parameters and identify the importance of them. Two well-known benchmark functions are used for evaluation purposes. Experimental results revealed that most parameters and their low-order interactions have a significant influence on the performance of the iABC. Furthermore, the proposed iABC proved its superiority compared to other state-of-the-art algorithms.