Classification of Sleep Apneas using Decision Tree Classifier

Abstract

In today's world, obstructive sleep apneas are the main reason for serious health problems. These sleep apnea disorders can have serious and life-shortening consequences. Obstructive sleep apnea (OSA) is a typical sleep problem brought about by unusual breathing. Polysomnography (PSG) is a traditional and highest quality level for OSA analysis to detect various sleep apneas more than 6 hours (approximately) recordings are needed which are very long-time recordings. As a human, it is difficult to identify the disorder from the ECG signals. So automated computer-based analysis is required to detect the diseases as early as possible. Therefore, this work aims for the automated and computer-based detection and classification of sleep apneas. In this work, MIT-BIH Polysomnographic (18 ECG signals) are considered as input signals. DWT of 30 seconds epoch signal is utilized to extract the features. 7-level decomposition has been done by using the wavelet `sym3'. From each level, 12 features were extracted from all the beats to classify the five sleep apneas are; Normal (N), Hypopnea with arousal (HA), Obstructive apnea with arousal (X), Obstructive apnea (OA), and Central apnea with arousal (CAA). The experimental results show that the Decision Tree classifier achieves better performance with an average of 98.53% accuracy, 98.39% sensitivity, 96.86% specificity, 90% of precision, and an overall F-score of 93.2%.