An Optimized Obstructive Sleep Apnea Detection Model Using Particle Swarm Optimization and Machine Learning

Obstructive Sleep Apnea (OSA) is a severe health issue all over the world, characterized by repeated interruption in breathing during sleep. Traditional modes of diagnosis, like polysomnography, require huge costs and time for evaluation, due to which there is a need for efficient and automated diagnostic systems. This paper presents an enhanced model named Optimized Intelligent System for Obstructive Sleep Apnea (OISOSA) for efficient OSA detection using single-lead EEG data. The proposed model utilizes the discrete wavelet transform with db8 for the decomposition of the EEG signal into the sub-bands and extraction of features from each of the sub-bands. Further, the model incorporates Gaussian filter for feature smoothing, Isolation Forest algorithm for the removal of outliers, and Particle Swarm Optimization (PSO) algorithm for feature optimization. Finally, the classification of apnea and non-apnea is done using the extra tree classifier. Performance evaluation using metrics such as accuracy, precision, recall and F1-score have been used to demonstrate the effectiveness of the proposed system. OISOSA achieved the highest accuracy of 86% with 10-fold cross-validation and 85% with holdout validation, outperforming the benchmark models. This study highlights the potential of PSO-enhanced ensemble analytics for reliable and accessible OSA detection.