Obstructive sleep apnea prediction using deep learning

Sleep apnea is defined as the cessation of breathing for at least ten seconds and known as a common sleep disorder. Obstructive sleep apnea (OSA) is the most common type of sleep apnea that occurs due to obstruction in the airway. Besides the detection of sleep apnea with the help of developed algorithms, early prediction of this syndrome is important in order to prevent serious health problems and life-threatening situations. With the prediction models working with high accuracy, it will be possible to prevent the possible risks without experiencing them by stimulating the OSA patients before the syndrome occurs and waking them up from sleep. In this thesis, models that predict apnea using electrocardiographic signals of patients diagnosed with obstructive slep apnea by using Convolutional Neural Networks (CNN) are presented. The first model is the training from scratch with pre-trained architectures. The second model is the study of using the transfer learning method with pre-trained architectures. The third model is the study in which the new results observed with the changes made on the architecture that performed the best prediction performance in the first two models. In the last model, the results are presented when the features obtained from the deep learning architecture proposed in the third model are classified by Support Vector Machines, Random Subspace k-Nearest Neighborhood and Random Subspace Discriminant Analysis methods instead of the architecture's own classifier. The high accuracy findings observed at the end of the study show that the proposed model can be used as a good indicator for the prediction of OSA syndrome.