Sleep stage classification of sleep apnea patients using decision-tree-based support vector machines based on ECG parameters

This paper describes the design and validation of an effective sleep stage classification strategy for patients with sleep apnea. This strategy consists of a sequential forward selection (SFS) feature selection method and a decision-tree-based support vector machines (DTB-SVM) classifier for discriminating three types of sleep based on electrocardiogram (ECG) signals. Each 5-minute epoch of ECG signal data collected during sleep was used to generate 24 features using heart rate variability (HRV) analysis. An SFS feature selection method was then employed to determine which significant features should be selected to improve classification accuracy. A DTB-SVM was then trained using selected features in order to discriminate three sleep stages, including pre-sleep wakefulness, NREM sleep and REM sleep. The average classification accuracy of the proposed strategy was 73.51%. Our experimental results demonstrate that the proposed strategy provides moderate accuracy for detecting sleep stages in sleep apnea patients and can serve as a convenient tool for assessing sleep quality.