Machine Learning Based Sleep Apnea Detection Using EEG Signals

Sleep apnea is an acute and widespread sleeping disorder that impacts a person's breathing repeatedly when it occurs in 3 forms - Obstructive Sleep Apnea (OSA), Central Sleep Apnea (CSA) and Complex or Mixed Sleep Apnea. To manage this condition, it is essential to diagnose and identify it correctly. In this work, we explore how split band electroencephalogram (EEG) data can be used to correctly identify sleep apnea events using different Machine Learning (ML) models, such as Naive Bayes (NB), Random Forest (RF), K-Nearest Neighbours (KNN), and Decision Tree (DT). Our approach involves splitting the input EEG signals into frames and sub-frames which is then passed through a band-pass filter to extract suitable information from the data. These ML models are evaluated based on their accuracy, specificity, and F1-score. Our experimental results show that NB has consistently outperformed the other models in terms of performance indicators. Notably, the NB model achieves the highest accuracy in the Alpha band and the C3-A2 channel. Our findings suggest that certain E EG bands a nd channels provide more discriminative information for apnea detection, which could inform the development of advanced diagnostic tools.