Ensembled Obstructive Sleep Apnea Detection Using Extra Tree Ensemble Technique

Obstructive Sleep Apnea (OSA) is a sleep-related disorder with repetitive episodes of complete or partial upper airway obstruction during sleep. These episodes result in disrupted breathing patterns and reduced oxygen levels in the blood. Electroencephalogram is a valuable tool in detecting and diagnosing sleep apnea as it provides insights into the disruptions of sleep caused by repeated apnea events. This paper proposes an efficient model to detect OSA named Ensembled Expert System for Obstructive Sleep Apnea Detection (EESOSAD) using EEG physiological sensors and an Extra Tree ensemble classifier. The proposed model incorporates Discrete Wavelet Transform with Daubechies order-8 for decomposing EEG signal into multiple sub-bands and extracting statistical features like - Maximum, Minimum, Mean, Standard Deviation, Kurtosis, Power, Energy, Root Mean Square, and Variance. A state-of-the-art data preprocessing pipeline is also designed and developed for treating corrupted, missing, duplicates, and outliers, as medical data are prone to noise, anomalies, and inconsistencies that can distort patterns and compromise analysis validity. Finally, for the classification task, the Extra Tree ensemble classifier is used for OSA detection to comprehensively evaluate the performance of the developed data pre-processing and feature fusion pipeline. Multiple evaluation metrics, including Precision, Recall, Accuracy, F1-score, and AUC-ROC curve, have been employed to compare and assess the model’s effectiveness. Furthermore, diverse state-of-the-art ensemble learning models have been incorporated for benchmark comparison and discussion. The proposed EESOSAD model achieved an average accuracy of 59% and outperformed multiple benchmark models with an optimal performance margin.