A modified bees algorithm and a statistics-based method for tuning its parameters

This paper presents a novel version of the bees algorithm. This version is characterized by an extended set of search operators, and a mechanism that protects the most recently generated solutions from competition with more evolved individuals. Compared to the standard implementation of the bees algorithm, the new procedure requires the selection of an additional set of parameters. A new statistical method is proposed to tune these extra parameters. The proposed tuning method was used to determine a unique set of learning parameters for the modified bees algorithm on eight popular function optimization benchmarks. When tested against the standard bees algorithm and two other well-known optimization procedures, the new algorithm attained top performances on nearly all the benchmarks. The experimental results also proved that, tested on a search space much larger than that where it was tuned, the modified bees algorithm still outperformed the standard method, and the degradation of the performance of the two algorithms was comparable. These results prove the effectiveness of the modified bees algorithm, and show that the proposed tuning procedure is a valuable alternative to the complex and subjective trial-and-error methods that are often used.