Electrocardiogram-Based Feature Extraction for Machine Learning Classification of Obstructive Sleep Apnea

This paper introduces a new feature extraction technique based on Time Sequence Analysis, combined with machine learning classification technique called Extreme Learning Machine (ELM), for automatic diagnosis of Obstructive Sleep Apnea (OSA) syndrome. The feature was extracted from Electrocardiogram (ECG) signal of patients with and without OSA. The ECG recordings were labelled as “Apnea” or “Normal” by experts’ examination. The data was freely available online from Physionet database. The feature extraction and classification algorithms were implemented on Matlab environment and the performance was evaluated in terms of OSA detection accuracy percentage. The aim of the study is to provide a low computational feature extraction technique for automatic OSA diagnosis. Simulation results show that OSA detection with 80.3% accuracy is possible using one feature only. It is concluded that the proposed technique offers OSA diagnosis with good enough OSA detection while reducing