Abstract:
Sleeping is an important activity to monitor since it has a crucial role in the over all health and well-being of the people and the society. The problems in sleep affect the daily lives of people negatively and a great deal of diseases has a strong correlation with low sleep quality. In order to diagnose the problems in the sleep, different monitoring systems are developed in the literature. The unobtrusiveness, reduced cost and reach ability are the main design considerations and in this thesis, those are accomplished with the smart wearables; smart watch and smart phone. The proposed system can be utilized as a prescreening tool which recognizes the severity of problems in respiration during sleep. For this purpose, the accelerometer and heart rate monitor sensors on smart watch and the sound level sensor on smart phone are activated. The experiments of this system are performed with 17 subjects in a sleep clinic. The subjects are ex amined by the specialist doctor and among those subjects; nine of them perform a few abnormal respiratory events whereas lots of abnormal respiratory events are observed for remaining eight subjects. The data collected from these subjects is merged and used to generate various combinations by employing varied feature extraction, feature selection and sampling approaches. Five different machine learning algorithms are im plemented and the classification results are generated with the various combinations of data, training and scoring strategies. The system performance is measured in two ways; discrimination success of abnormal respiratory events and classification success of subjects according to the problems in their respiration. The best achieved accuracy rate of distinguishing abnormal respiratory events is 85.95% whereas the classification success of subjects is one misclassification through 17 subjects.
