Assessment of Recurrence Quantification Analysis (RQA) of EEG for Development of a Novel Drowsiness Detection System

Purpose In this paper, the aim is to detect drowsiness using one of the well-known nonlinear signal analysis methods known as Recurrence Quantification Analysis (RQA). We want to show that by assuming brain as a chaotic system, the number of recurrences in the phase space of this system will increase during drowsiness state. Methods Determinism (DET) feature extracted by Recurrence quantification analysis (RQA) method has been used to detect these recurrences. Furthermore, eleven other features of RQA for the purpose of comparing their capability with DET feature have been used to detect drowsiness. Three different feature subsets are extracted from these twelve features. The first feature subset is called DET feature. The second feature subset is obtained by applying Linear Discriminant Analysis (LDA) technique on the twelve dimensional feature set. The third feature subset is made by Sequential Forward Selection (SFS) method. To reach the highest value of accuracy, specificity and sensitivity, the three evaluated feature sets have been applied to four different classifiers known as K-nearest neighbor (KNN), Support Vector Machine classifier (SVM), Naive Bayes and Fisher Linear Discriminant Analysis. A K-means clustering method has also been applied on the data to ensure that the criteria used for labeling drowsy and alert segments are suitable. Results The Results reveal that DET feature could achieve the best performance in drowsiness detection by SVM classifier with an accuracy of more than ninety percentage. Conclusions These findings approve that DET measure is a reasonable feature for the purpose of drowsiness detection.