The Enhanced Mayfly Optimization Algorithm with Roulette Wheel Selection

Abstract

In the year 2020, the Mayfly optimization method was proposed. It is a modification of particle swarm optimization and it combines major advantages of particle swarm optimization, genetic algorithm, and firefly algorithm. Mayfly flight and mating activity were the inspiration for this piece. Simulated in many tests using various benchmark functions, all of which were found to be capable of optimization, although some drawbacks, like a sluggish or premature convergent rate, and a probable imbalance between exploration and exploitation, have
yet to be handled, necessitating modification for improved performance. The Mayfly Algorithm hasn't been used much for feature selection problems, to the author's knowledge. In this study, the Mayfly algorithm was enhanced with the Roulette Wheel Selection method been the most common and straightforward method of fitness proportionate selection, free of bias, because each individual is given a fair chance of selection, preserving diversity. On the constructed database, the evaluation is based on the force acceptance rate, force rejection rate, recognition accuracy, and recognition time. The created database is mainly for purpose of this study. Five hundred
and seventy images (570) of face and iris were acquired via digital camera, three hundred and forty-two (342) face and iris images were used for training which equals 60% of the total dataset and two hundred and twenty-eight (228) face and iris images which are equivalent to 40% of the total dataset were used for testing. Both unimodal and multimodal recognition systems were used in the stimulation trials. The optimal result was achieved on a fused recognition system at a threshold of 0.76. The findings reveal a 1.79% force acceptance rate, 2.92% force rejection rate, 97.36% recognition accuracy, and 181.52 sec recognition time for enhanced Mayfly algorithm (EMA) as against 3.51% force acceptance rate, 5.26% force rejection rate, 95.18% recognition accuracy, and 215.75 sec recognition time for original Mayfly algorithm (MA). Obtained results showed that the enhanced algorithm would indeed increase the capability of the original Mayfly algorithm.