Hyperparameter tuning of convolutional neural networks using nature-inspired metaheuristic algorithms for image classification
Naveen Kumar Baskaran, Bhavy Pratap, Sulabh Bansal · 2025
This chapter explores how optimization techniques can enhance image classification models, focusing on three bioinspired algorithms: Genetic Algorithms (GAs), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). These algorithms help fine-tune hyperparameters, improving a model’s ability to extract features and recognize patterns more effectively. We begin with an overview of optimization and its importance in image classification. The chapter then delves into how GA, PSO, and ACO work, explaining their unique mechanisms. To test their effectiveness, we applied them to a Convolutional Neural Network (CNN) trained on the Fashion-MNIST (Modified National Institute of Standards and Technology) dataset. The results showed that optimized CNNs achieved better accuracy, faster convergence, and improved computational efficiency. By fine-tuning CNN parameters, these optimization techniques significantly boosted precision and recall, making the model more robust. Our findings highlight the potential of bioinspired algorithms to enhance deep learning models, paving the way for future research in hyperparameter tuning, exploring other optimization methods, and applying these techniques to more complex datasets.