Intelligent Sleep Apnea Detection by Advanced ML Using Single-Lead EEG Signal Data

Obstructive sleep apnea (OSA) is a common sleep disorder that causes repeated disruptions in breathing during sleep. The traditional method to diagnose OSA is polysomnography, which is complex and time-consuming and requires an overnight stay in a sleep lab. However, electroencephalography (EEG) sensor-based methods offer the possibility of simpler, in-home testing, improving accessibility and patient comfort. This article proposes an intelligent expert system for the OSA detection (IESOSAD) model. The proposed model aims to efficiently detect apnea utilizing single-lead EEG data and ensemble learning algorithms. The IESOSAD model begins by analyzing the C4-A1 channel of the EEG signal and uses discrete wavelet transform (DWT) with a Daubechies-8 wavelet (db8) to decompose it into subbands. Statistical features (Sfs) are then extracted from these subbands to create a dataset for further analysis. Furthermore, the dataset undergoes preprocessing with a Gaussian filter for feature smoothing and isolation forest (IF) for anomaly detection, leading to enhanced data quality. Subsequently, the artificial bee colony (ABC) feature selection algorithm is applied to eliminate irrelevant features. The final stage of the IESOSAD model involves classification using an Extremely Randomized Trees classifier. The IESOSAD model’s performance is rigorously evaluated under holdout, tenfold, and fivefold cross-validation (CV) using a comprehensive set of metrics, including precision, recall, accuracy, ${F}1$ -score, and ROC AUC curve. The results demonstrate that IESOSAD achieves the highest accuracy of 88.12%, surpassing other state-of-the-art machine and ensemble learning algorithms. Moreover, IESOSAD has outperformed benchmark OSA detection models by a significant margin, facilitating a more streamlined and reliable OSA detection system.