ESAD: Expert System for Apnea Detection Using Enhanced DWT Feature Extraction and Machine Learning Algorithms

Sleep apnea is a sleep disorder with some potentially severe consequences, characterized by recurring disturbance in breathing during sleep. This happens when the muscles in the back of the throat cannot maintain an open airway, leading to brief awakenings to take a breath. Since the last decades, several expert systems have been proposed for apnea detection, but only a few have analyzed the data exhaustively before classification. This paper proposes a model for apnea detection named Expert System for Apnea Detection (ESAD) that uses Discrete Wavelet Transform (DWT) for EEG signal decomposition and for features extraction, namely Standard Deviation(SD), Power, Mean, Energy, Kurtosis, Max, Min, Variance and Root Mean Square (RMS). The proposed ESAD uses Isolation Forest (IF) for outlier detection as the used medical data can have noise and anomalies. Finally, the SVM with Radial Basis Function (RBF) kernel is incorporated for detecting and classifying sleep apnea disorder accurately. The model’s performance is evaluated by incorporating key performance metrics like - Precision, Accuracy, F1-Score and Recall and ROC-AUC curve. On comparing ESAD with other machine learning models, it outperforms other incorporated shallow learning models like - Decision Tree, KNN, and Naive Bayes with superior performance, demonstrating higher accuracy and precision in detecting apnea events.