Hyperparameter Optimization of Deep Learning Model: A Case Study of COVID-19 Diagnosis

The global impact of COVID-19, which has affected over 700 million individuals, necessitates the development of automated diagnostic tools for rapid screening using clinical imaging, such as X-rays. Deep learning has shown remarkable capabilities in feature extraction and classification, making it a promising technique for automatic diagnosis of COVID-19 cases through analysis of chest X-ray (CXR) images. However, achieving optimal classification performance with deep learning models relies heavily on properly setting the hyperparameters during the transfer learning process, presenting a nontrivial challenge. This paper introduces sperm swarm optimization (SSO), an emerging metaheuristic search algorithm, for fine-tuning four key hyperparameters of convolutional neural networks (CNNs) to ensure effective training of the network. The proposed model, SSOCNN, is evaluated using a publicly available database comprising CXR images with normal, pneumonia, and COVID-19 cases. Our results demonstrate the promising performance of SSOCNN in automatic diagnosis of COVID-19 cases, achieving accuracy, sensitivity, specificity, precision, and F1 score values of 96.54%, 97.41%, 98.52%, 97.05%, and 97.23%, respectively.