TSmRMR Feature Selection Method for Detecting Sleep Apnea Episodes Based on EEG Signals

• Published in: IEEE-EMBS Conference on Biomedical Engineering and Sciences pp. 103 - 107
• Main Authors: Hsieh, Chia-Tung, Huang, Yuan-Hao
• Format: Conference Proceeding
• Language: English
• Published: IEEE 11.12.2024
• Subjects: Accuracy; Classification algorithms; EEG signals; Electroencephalography; Feature extraction; machine learning; mRMR algorithm; Nearest neighbor methods; Redundancy; Search methods; Sleep apnea; sleep apnea-hypopnea syndrome; Time-frequency analysis; Training
• ISSN: 2573-3028
• DOI: 10.1109/IECBES61011.2024.10991314

Electroencephalogram (EEG) signals are extensively used in sleep-related research, including detecting sleep apnea-hypopnea syndrome (SAHS), characterized by frequent breathing interruptions during sleep. However, extracting and selecting effective features from EEG signals for SAHS detection poses significant challenges. This paper presents a novel feature selection method that integrates the threshold search method with the widely used maximal relevance minimal redundancy (mRMR) algorithm. To distinguish between normal, hypopnea, and apnea episodes in SAHS subjects, various time-frequency features were extracted from EEG signals and classified using a K-nearest neighbor (KNN) classifier. Compared to the benchmark mRMR algorithm, the proposed threshold search mRMR (TSmRMR) algorithm reduced the number of features from 27 to 17. The classification results showed that the TSmRMR achieved an accuracy of 96.67%, representing a 2.68% improvement over the mRMR algorithm.