Adaptive and Extensive Classification for Optimal Multi-Feature Selection on Covid's Data

Abstract

Coronavirus (CD) is a disease that impacts the brain's motor system. Tremors, muscular rigidity, inaccurate gait, and speech difficulty are the hallmarks of CD. A definitive diagnosis of Covid's disease generally requires multiple neurological, psychological, and physical examinations, even though the main symptoms cannot be readily distinguished from those of other disorders. This has led to a number of recent initiatives to use machine learning-based automated diagnostic assistance systems to better assess CD patients. Automatic early identification of COVID-19 using feature data sets is one of the most difficult medical problems that exist today. Some features of these datasets are useless, and others are rife with problems like noise that make learning difficult and increase the amount of computing power needed. To improve the efficiency of feature selection and guarantee the classifier's most accurate performance. Implement a Novel Stacked Convolutional Neural Network Model (NSCNNM) in this study to automatically diagnose COVID-19 illness.Our objective is to develop a system, which can accurately detect and categorize COVID-19 using chest radiographs and is based on deep learning. We begin by contrasting the efficiency of different cutting-edge convolutional neural networks (CNNs) suggested for medical picture categorization in the past few years. Second, when building and developing CNN, we begin at the very beginning. In both situations, we train and validate with the publicly accessible X-Ray dataset. We accomplished the tertiary classification task (Normal/COVID-19/Pneumonia) with 87.50% correctness and the binary classification task (Normal/COVID-19) with 98.33% accuracy by means of transfer learning. The CNN that was trained from the ground up achieves an accuracy of 93.75 percent when it comes to tertiary classification. The accuracy of classification using transfer learning drops as the number of classes grows. The results are shown by a 10-fold cross-validation confusion metric study and comprehensive receiver operating characteristics (ROC) analysis.