<![CDATA[Driver drowsiness is a critical road safety issue responsible for thousands of traffic fatalities annually, particularly on monotonous road environments such as toll highways. This research introduces an Electroencephalography (EEG)-based driver drowsiness classification model using a Support Vector Machine (SVM) algorithm, with Observer Rated Sleepiness (ORS) as the ground truth. Data were collected from 22 licensed drivers who completed a 90-minute monotonic driving simulation in an ergonomics laboratory using the Muse S portable EEG device. The features extracted from the EEG consisted of mean and standard deviation for delta, theta, alpha, and beta band power each (14 in total) per one-minute epoch, as well as relative band power for each band and two ratios: the theta/alpha (?/?) ratio and (theta+alpha)/beta ((?+?)/?) ratio. Each classification was carried out using an SVM with a Radial Basis Function (RBF) kernel, and a One vs Rest (OvR) multi-class strategy, and the generalization power was validated using Leave-One-Subject-Out Cross Validation (LOSO-CV), which exhibits subject-independent generalization. Results showed an average LOSO-CV accuracy of 71.97% (SD = 14.67%) with a Macro-F1 score of 0.7187. The ?/? ratio was the most discriminative feature, increasing from ?0.54 (alert) to 1.59 (severely drowsy).]]>
