Citation

Fouad IA. Ain Shams Eng. J. 2023; 14(3): e101895.

Copyright

(Copyright © 2023, Ain Shams University, Publisher Elsevier Publishing)

DOI

10.1016/j.asej.2022.101895

PMID

unavailable

Abstract

Vehicle accidents on long routes around the world are frequently caused by drowsy drivers. It is mainly because there is no system that measures alertness. The driver will be notified to interrupt his/her travel if an accurate and robust fatigue detection system is available. Dealing with this approach will help the driver avoid accidents and make the right decisions. This paper aims to detect drivers' sleepiness using a powerful software tool. It was initially developed by capturing electroencephalography (EEG) signals and processing them. In this research, different machine learning algorithms were applied to the EEG signals of twelve subjects to measure their performance. In the first step, all recorded data for all subjects were segmented into second epochs. Brain signals were labeled alert or drowsy for each epoch. Before applying the machine learning algorithms to the epoched signal, a preprocessing step is introduced to extract the relevant features. The applied algorithms are: Naive Bayes (Diagonal Linear Discriminant Analysis), Support Vector Machines (Linear and Radial Basis Functions), K-Nearest Neighbor (KNN), and Random Forest Analysis (RFA). By capturing signals from only three electrodes, it was found that utilizing more than one classifier led to the highest accuracy of 100% for all subjects considered in this study. In general, this developed EEG-based system detects drowsiness and loss of focus of drivers in real-time with high accuracy, making it a practicable and reliable option for real-time applications.

Language: en

Keywords

Drowsiness detection; Electroencephalography (EEG); K-Nearest Neighbor (KNN); Linear discriminant analysis (LDA); Random forest (RF); Support vector machine -RBF (SVM-rbf)