Classification of Sleep Apnea Event Type Using Imbalanced Labelled EEG Signal

Sleep apnea is a breathing disorder due to short time breath interruption during sleep. Patients suffering from this disorder exhibit symptoms like day time fatigue, drowsiness etc. and even lead to severe cardiovascular problems. This disorder is treated based on sleep apnea event types encoun-tered. Therefore, classification of sleep apnea event types is an important and essential part for patient's treatment. Several machine learning (ML) approaches have been developed to address this problem with Electroencephalogram (EEG) signal. However, most of these EEG signals are annotated in imbal-anced form which introduces inaccurate classification of event types. This paper proposes an approach to handle imbalance EEG signal annotations for achieving more accurate sleep apnea event types classification. For this purpose, minority samples are oversampled by using ADASYN technique and pairs of highly similar samples belonging to different classes are removed with the help of Tomek-links method. The proposed approach has been experimented on MIT-BIH Polysomnographic Database and St. Vincent's University Hospital/University College Dublin Sleep Apnea Database (UCDDB). Four widely used classifiers such as bagged trees, subspace KNN, KNN and cubic SVM are considered in this study to test the performance of the proposed method, out of which, subspace KNN achieved highest accu-racy of 98.92% and 81.39% on MIT-BIH Polysomnographic Database and UCDDB database respectively.