Abstract

Driver drowsiness is one of the leading causes of fatal road traffic accidents (RTA). While studies have illustrated the effectiveness of spatial criteria on driver drowsiness, the effects of these factors on drowsy driver behavior have yet to be mapped or spatially modeled properly. Meanwhile, driver behavior has been known as a major factor in the RTA occurrence. This paper investigated the effect of four spatial criteria of chainage, distance traveled after a horizontal curve, horizontal curvature, and slope on drowsy driver behavior in a simulated driving task. Accordingly, the drowsiness level on a monotone 108 km virtual expressway in northeastern Iran was determined by modeling the EEG data of 20 subjects. Four machine learning (ML) algorithms, namely bagged decision trees (BDT), multi-layer perceptron (MLP), random forest (RF), and support vector regression (SVR), were used to analyze electroencephalogram (EEG) signals and to estimate driver drowsiness based on spatial criteria. The results of the receiver operating characteristic curve (ROC) showed that BDT (0.925), RF (0.919), MLP (0.807), and SVR (0.706) had the highest accuracy in estimating drowsiness level on the virtual road using EEG signals, respectively. Using the estimated drowsiness level by BDT and spatial criteria values, ML algorithms were trained to map driver drowsiness on the Tehran-Mashhad road. Based on ROC results, MLP (0.921), BDT (0.918), RF (0.917), and SVR (0.896) had the most accurate performance in mapping driver drowsiness, respectively. Generated drowsiness level maps were validated, and the estimations of BDT and RF were more consistent with real drowsy driving accident black spots in the study area. In conclusion, mapping driver drowsiness by applying ML algorithms and spatial criteria detects high-risk areas and helps to improve road safety.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call