<p>Neural Architecture Search (NAS) automates the design of Convolutional Neural Network (CNN) architectures, reducing the need for manual intervention and allowing models to perform more effectively across different tasks and datasets. However, the rapid growth of deep learning has created a strong demand for more efficient and scalable architecture design methods. Conventional NAS methods often search for the optimal architecture by testing all possible combinations and configurations, which is time-consuming and computationally expensive. Therefore, various research efforts have been conducted to optimize this technique. The best techniques should ensure better performance by balancing model complexity, accuracy, and computational efficiency. Despite the availability of various optimization techniques, there is a gap in the literature regarding the integration of advanced optimization methods for NAS to further reduce the computational complexity and improve the efficiency of the search process. In this paper, we integrate Particle Swarm Optimization (PSO) with the Iterated Local Search (ILS) algorithm to optimize efficiently the NAS. Our proposed algorithm combines the global exploration capabilities of PSO with the local refinement capabilities of ILS. We trained and conducted experiments on three prominent image datasets, MNIST, MNIST with background images (MBI), and the rectangles dataset (RECT). Furthermore, we tested and evaluated the performance of our proposed algorithm on the COVID-19 dataset. Our results demonstrate that the PSO-ILS algorithm consistently outperforms a state-of-the-art algorithm (IDECNN) in terms of classification accuracy across datasets, achieving 95.65%, 95.87%, and 97.46%. Further analysis reveals promising trends in performance metrics such as precision, recall, and F1-score, highlighting the effectiveness of our proposed solution.</p>