A modified version of the ABC algorithm and evaluation of its performance

The artificial bee colony (ABC) optimization algorithm has been widely used to solve the global optimization problems. Many versions of ABC algorithm exist in the literature intending to achieve optimum solution for problems in different domains. Some modifications of the ABC algorithm are general and can be applied to any problem domain, while some are application dependent. This paper proposes a modified version of the ABC algorithm named as, MABC-SS (modified artificial bee colony algorithm with selection strategy), that can be applied to any problem domain. The algorithm is modified in terms of population initialization and update of a bee position using the old and a new food source equation based on the algorithm's performance in the previous iteration. The selection strategy is measured based on a novel approach called the rate of change. The population initialization in any optimization algorithm plays an important role in achieving the global optimum. The algorithm proposed in the paper uses random and an opposition-based learning technique to initialize the population and updates a bee's position after exceeding a certain number of trial limits. The rate of change is based on the average cost and is calculated for the past two iterations and compared for a method to be used in the current iteration to achieve the best result. The proposed algorithm is experimented with 35 benchmark test functions and 10 real world test functions. The findings indicate that the proposed algorithm is able to achieve the optimal result in most cases. The proposed algorithm is compared with the original ABC algorithm, modified versions of the ABC algorithm, and other algorithms in the literature using the test mentioned above. The parameters such as population size, number of iterations and runs were kept same for comparison with non-variants of ABC. In case of ABC variants, ABC specific parameters such as abandonment limit factor (0.6) and acceleration coefficient (1) were kept same. Results indicate that in 40% of the traditional benchmark test functions, the suggested algorithm works better than other variants of ABC (ABC, GABC, MABC, MEABC, BABC, and KFABC), while 30% of the traditional benchmark test functions are comparable. The proposed algorithm was compared to non-variants of ABC as well. The results show that the proposed algorithm achieved the best mean result in 50% of the CEC2019 benchmark test functions and in 94% of the classical benchmark test functions. The result is confirmed by Wilcoxon sum ranked test which shows that MABC-SS achieved statistically significant result in 48% of the classical and 70% of the CEC2019 benchmark test functions when compared with the original ABC. Overall, based on assessment and comparison in benchmark test functions used in this paper, the suggested algorithm is superior to others.