A neural network based approach for COVID-19 detection using jellyfish-optimized feature selection

Various methods have been proposed to COVID-19 detection from healthy people. In many of these methods, computed tomography (CT) scans of COVID-19 have been used for classification. Here, it is assumed that COVID-19 already has affected and damaged the lung tissue of patients which remains the main pitfall of these methods. Studies show that in many cases, a person infected with COVID-19 has healthy lungs, so there is a need to develop methods to intelligently classify COVID-19 and healthy people based on other clinical information. In this paper, a new approach uses a classification algorithm based on a multilayer Artificial Neural Network (ANN) combined with a Jellyfish Search Algorithm (JSA) for classifying COVID-19 patients and healthy people in order to optimize multilayer ANN parameters and selecting data features is presented. The proposed JSA+MLP (Multi-Layer Perceptron) method has been evaluated over five independent runs with stratified 70/15/15 train/validation/test splits. Results demonstrate 96.44% accuracy, 96.38% precision, and 96.54% sensitivity, and 96.46% F1 Score, outperforming ANN, CNN, LSTM, RNN, CNN-LSTM, CNN-RNN baselines, and three alternative meta-heuristic feature selectors. Convergence, stability, and Pearson correlation analyses are reported across all runs for full reproducibility.