Mean Teacher Model with Consistency Regularization for Semi-supervised Detection of COVID-19 Using Cough Recordings

This study provides a novel approach to detecting COVID-19 using cough recordings. The approach uses labeled and unlabeled data and a Mean Teacher model with consistency regularization. A student network and a teacher network make up the model, and the teacher network directs the student network’s training. The consistency of the student network’s predictions is preserved, which enhances generalization. The recordings used were all mixed labeled and unlabeled coughs of COVID-19 patients and healthy people. Using k-fold cross-validation, accuracy, loss, recall, precision, and AUC are all measured. The outcomes imply that COVID-19 identification under semi-supervised conditions is feasible. The results highlight the importance of cough samples in the absence of labeled data.