Performance Analysis of Metaheuristic Algorithms on Benchmark Functions

Abstract

The discipline of optimization can be used to maximize or minimize several problems. The use of metaheuristic algorithms is a strategy that often works well for global optimization. They are a type of stochastic algorithm that, via trial and error, finds workable solutions to difficult optimization problems in a reasonable amount of time, but they do not provide assurance that the answers are optimal. This paper aims to offer a comparative analysis of several metaheuristics in searching for the optimal solution. The selected metaheuristics are Artificial Bee Colony, Ant Lion Optimizer, Bat, Black Hole, Cuckoo Search, Cat Swarm Optimization, Dragonfly, Differential Evolution, Firefly, Genetic, Gravitational-Based Search, Grasshopper Optimization, Grey Wolf Optimizer, Harmony Search, Krill-Herd, Moth-Flame Optimizer, Particle Swarm Optimization, Sine Cosine, Shuffled Frog-Leaping, and Whale Optimization algorithms. For this evaluation, 18 benchmark test functions, categorized as unimodal, multimodal, and fixed-dimension multimodal are used to examine various properties, such as accuracy, escape from the local optimum, and convergence. As an indicator of how effectively these metaheuristics work, metrics like minimum, maximum, average, and standard deviation of fitness are provided. There are no optimization algorithms that are adequate for all problems, as the No Free Lunch theorem suggests, but the metaheuristics that are more effective than the others will be demonstrated. This study could be helpful for young researchers to identify the most prominent metaheuristics for achieving a better global optimum.

Received: 8 June 2024 / Accepted: 25 July 2024 / Published: 29 July 2024Optimization, Metaheuristic algorithm, Benchmark function, Performance metric