OCRUN: An oppositional Runge Kutta optimizer with cuckoo search for global optimization and feature selection

Abstract

The recently proposed swarm intelligence algorithm, Runge–Kutta Optimization (RUN), is rooted in the fourth-order Runge–Kutta method. Compared with its counterparts, RUN boasts an advantage of having a more concrete theoretical foundation embodying a more powerful optimization efficacy, free from any metaphor. However, RUN still has its shortcomings. The compelling enhanced solution function leads to insufficient exploration ability of the algorithm, and resulting in an imbalance between exploration and exploitation that cannot be mitigated. An improved version based on opposition-based learning and cuckoo search is proposed to compensate for the above deficiencies, called OCRUN. OCRUN is tested on 30 test functions of CEC2014 with 10 classical metaheuristics and 9 advanced metaheuristics, respectively. Combining the experimental results and the Wilcoxon signed-rank test, OCRUN exhibits excellent performance. At the same time, parameter sensitivity analysis experiments are also carried out on this test set. Furthermore, a binary implementation of the algorithm was constructed specifically for feature selection cases, labeled as BOCRUN. BOCRUN is compared with 5 existing binary metaheuristics on 15 public datasets. The experimental results show that the improved algorithm performs well in feature selection. Therefore, OCRUN is an effectively improved optimizer. Finally, the OCRUN method offers high-quality solutions to engineering problems and contributes significantly to engineering design under practical constraints. The method has been successfully applied to various design scenarios, such as reducer design, cantilever beam design, and tension/compression spring design. The OCRUN method outperforms other similar products in terms of performance and effectiveness.