TY - GEN
T1 - Deep Learning for Single-Channel EEG-Based Driver Drowsiness
T2 - 23rd International Arab Conference on Information Technology, ACIT 2022
AU - Latreche, Imene
AU - Slatnia, Sihem
AU - Kazar, Okba
AU - Harous, Saad
AU - Athamena, Belkacem
AU - Houhamdi, Zina
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - To prevent car accidents caused by drowsiness, a transitional state between alertness and sleepiness, numerous authors have conducted studies employing various monitoring approaches and detecting techniques. However, the Electroencephalogram (EEG) was the most used monitoring approach due to its advantages compared to other techniques. Also, Deep Learning (DL) techniques have been widely used in this context and have given prominent results. In this paper, we have established a comparative study between four deep learning methods. The aim of this comparison is to determine the optimal DL model for detecting EEG-Based Drowsiness using a small balanced dataset. We evaluated the performance of the models using the hold-out method and the Leave-One-Subject-Out Cross-Validation (LOSO CV) techniques. With an accuracy of 77% with the Hold-Out method and 70.31% with the LOSO CV method, the GRU model outperforms other deep learning and machine learning models, followed by the BI-LSTM with 74.88% and 68.07, respectively.
AB - To prevent car accidents caused by drowsiness, a transitional state between alertness and sleepiness, numerous authors have conducted studies employing various monitoring approaches and detecting techniques. However, the Electroencephalogram (EEG) was the most used monitoring approach due to its advantages compared to other techniques. Also, Deep Learning (DL) techniques have been widely used in this context and have given prominent results. In this paper, we have established a comparative study between four deep learning methods. The aim of this comparison is to determine the optimal DL model for detecting EEG-Based Drowsiness using a small balanced dataset. We evaluated the performance of the models using the hold-out method and the Leave-One-Subject-Out Cross-Validation (LOSO CV) techniques. With an accuracy of 77% with the Hold-Out method and 70.31% with the LOSO CV method, the GRU model outperforms other deep learning and machine learning models, followed by the BI-LSTM with 74.88% and 68.07, respectively.
KW - Bi-LSTM
KW - CNN
KW - Cross-Subject
KW - Deep Learning
KW - Driver Drowsiness
KW - Electroencephalogram (EEG)
KW - GRU
KW - LSTM
UR - http://www.scopus.com/inward/record.url?scp=85146870625&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85146870625&partnerID=8YFLogxK
U2 - 10.1109/ACIT57182.2022.9994170
DO - 10.1109/ACIT57182.2022.9994170
M3 - Conference contribution
AN - SCOPUS:85146870625
T3 - Proceedings - 2022 23rd International Arab Conference on Information Technology, ACIT 2022
BT - Proceedings - 2022 23rd International Arab Conference on Information Technology, ACIT 2022
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 22 November 2022 through 24 November 2022
ER -