A learning automata-based hybrid MPA and JS algorithm for numerical optimization problems and its application on data clustering

摘要

Nature-inspired meta-heuristic algorithms possess various actions inspired by natural phenomena, animal behaviors, chemistry or physics laws etc. The actions are utilized to relocate agents in the problem space and update solutions. Selecting the best action to update a particle is a challenging responsibility in solving real-world NP-Hard problems. The selection process in meta-heuristic algorithms is predominantly random or quasi-random, which is not suitable enough. The Learning-Automata (LA) is a mechanism to make the selection process more dynamic. The LA chooses the most optimal action regarding history and status and balances exploration and exploitation intelligently. In the current paper, a neoteric LA-based hybrid optimization algorithm is presented for global optimization problems. In the proposed algorithm, the artificial Jellyfish search algorithm (JS) and Marine Predator Algorithm (MPA) are rectified to reduce their computational complexity while retaining their strengths. Furthermore, the LA’s probability vector is augmented to make it more efficient. Afterward, the JS and MPA algorithm are redeveloped using the proposed LA mechanism. The proposed LA-based hybrid is tested on thirty-eight low- and high-dimensional benchmark functions compared with state-of-the-art algorithms, statistically and visually. Moreover, proficiency of the proposed LA-based hybrid algorithm is investigated on the data clustering problem. The proposed algorithm is applied to ten datasets, and the results are compared with competitor algorithms using various metrics. The experimental results exposed the preponderance of the proposed LA-based hybrid algorithm.