Comparative Evaluation of Metaheuristic Algorithms for CNN Hyperparameter Optimization in Age-Invariant Face Recognition

Abstract

Age-invariant face recognition (AIFR) remains a challenging problem due to the significant facial appearance changes caused by aging, which often degrade the discriminative capability of conventional deep learning models. Although convolutional neural networks (CNNs) have demonstrated remarkable success in face recognition, their performance is highly dependent on the selection of optimal hyperparameters. This study presents a comprehensive comparative evaluation of six metaheuristic optimization algorithms which includes Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Nomadic Pastoralist Optimization Algorithm (NPOA), Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA) and Harris Hawks Optimization (HHO), for CNN hyperparameter optimization in age-invariant face recognition. The optimization process simultaneously tunes key CNN hyperparameters, including learning rate, convolutional filters, dropout rate and dense layer units, to enhance feature extraction and classification performance. The optimized CNN models were evaluated across multiple age groups using standard performance metrics, including accuracy, precision, recall, F1-score, convergence behaviour, computational time and age-group recognition accuracy. Experimental results demonstrate that all investigated metaheuristic algorithms significantly improve CNN performance over the initial configuration. Among them, PSO achieved the best overall performance with an accuracy of 98.56%, precision of 98.42%, recall of 98.47% and F1-score of 98.44%, while requiring the shortest training time (3.34 h) and the fewest convergence iterations (61). NPOA closely followed with an accuracy of 98.21%, demonstrating competitive optimization capability. Furthermore, PSO consistently produced the highest recognition rates across children, young adults, adults and elderly age groups, confirming its robustness to age-related facial variations.