ROMJIST Volume 23, No. 3, 2020, pp. 292-310
Prabhat Kumar Upadhyay, Chetna Nagpal Wavelet Based Performance Analysis of SVM and RBF Kernel for Classifying Stress Conditions of Sleep EEG
ABSTRACT: The aim of this study is to detect the changes in frequency and power through wavelet transform and assess the effects of externally induced heat stress in the classification of sleep EEG states - Awake, Slow Wave Sleep (SWS), and Rapid Eye Movement (REM). Radial Basis Function Neural Network (RBFNN) and Support Vector Machine (SVM) algorithms have been successfully applied to classify sleep stages under acute and chronic stress conditions with respect to their controlled Polymosograph. Performance analysis on test data set consists of performance metrics specificity, sensitivity and accuracy. The detection rate of SVM for sleep stage classification has been achieved with an average accuracy of 96.4% under chronic stress and 94.1% for acute stress, whereas an overall accuracy of 87% is achieved using RBFNN in classification of sleep-wake states. The results were also obtained for the identification of types of thermal stress i.e. acute and chronic using SVM approach with 84.4% detection rate. These results demonstrate that RBFNN and SVM models may be successfully employed in clinical studies as a decision support tools to confirm the presence of stress level.KEYWORDS: Wavelet transform; Heat stress; RBF; SVM; Sleep EEGRead full text (pdf)