Evaluation of different machine learning algorithms for classification of sleep apnea

Abstract

The syndrome of cessation of breathing with recurrent attacks for 10 seconds or more as a result of narrowing or obstruction of the upper respiratory tract is called sleep apnea (SA). As a result of not treating SA, serious problems such as hypertension, heart diseases, obesity and nervous disorders can occur. In recent years, studies of automatic diagnosis and prediction of SA have become popular. In this study, heart rate variability (HRV) signals were obtained using R peak information from from electrocardiography signals divided into one-minute segments. Time and frequency domain features were determined from HRV signals and apnea classification was made from the determined features by using five different machine learning algorithms. In this study, the highest accuracy was obtained from the Random Forest algorithm with 85.26%, the highest sensitivity was obtained from the K-Nearest Neighborhood algorithm with 78.08%, and the highest selectivity was obtained from the Random Forest algorithm with 91.4%.