QLESCA: A reinforcement learning based metaheuristic for efficient CNN filter pruning

Convolutional neural networks (CNNs) have demonstrated remarkable accuracy in detecting the COVID19 virus. However, their widespread adoption is hindered by their large size and high resource requirements, posing challenges for deployment on limited devices. To address this issue, filter pruning techniques have emerged as a promising approach to reduce the computational complexity of CNNs. In this study, we focus on the selection of the most effective filters in a CNN architecture utilizing the Q-Learning embedded SCA (QLESCA) as an optimizer. The effectiveness of QLESCA is evaluated by considering the recognition accuracy of a linear support vector machine (SVM). The proposed model undergoes training and testing on five sets of chest X-ray images, which consist of 4,848 COVID19 images and 8,669 non-COVID19 images. Numerical results demonstrate the superior performance of QLESCA, surpassing six recent metaheuristic optimization algorithms with a fitness value of −0.9781. By reducing the number of filters from 128 to 51, our proposed method achieves a mean accuracy of 97.222% on dataset 1, showcasing its efficacy in selecting a minimal set of filters. Moreover, we evaluate the best-detected model on dataset 1 with four additional datasets, yielding impressive accuracies of 97.847%, 97.935%, 97.695%, and 96.154% for datasets 2, 3, 4, and 5, respectively. This study highlights the challenge of deploying large CNNs on resource-limited devices and presents filter pruning as a solution. Leveraging the QLESCA optimizer, our approach effectively reduces the computational complexity of CNNs while maintaining high accuracy in COVID19 detection. The results demonstrate the efficiency and efficacy of our proposed method, showcasing its potential for practical deployment in healthcare settings.