Enhancing convolutional neural networks in electroencephalogram driver drowsiness detection using human inspired...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Transportation Safety

Background:

• Driver drowsiness is a major cause of traffic accidents.
• Electroencephalogram (EEG) signals are used for drowsiness detection.
• Deep neural networks (DNNs) are needed to analyze complex EEG data, but their optimization is challenging.

Purpose of the Study:

• To investigate the effectiveness of meta-heuristic algorithms for optimizing deep learning models in EEG-based driver drowsiness detection.
• To compare Teaching Learning Based Optimization (TLBO) and Student Psychology Based Optimization (SPBO) for optimizing Convolutional Neural Networks (CNNs).

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for EEG-based drowsiness detection.
• Employed Teaching Learning Based Optimization (TLBO) and Student Psychology Based Optimization (SPBO) algorithms to optimize CNN architectures.
• Evaluated model performance using metrics such as Area Under the Curve (AUC).

Main Results:

• Both CNN-TLBO and CNN-SPBO achieved strong predictive performance with AUC values of 0.926 and 0.920, respectively.
• TLBO resulted in a simpler CNN model (4,145 parameters) compared to SPBO (264,065 parameters).
• SPBO demonstrated faster optimization (116 minutes vs. 148 minutes for TLBO) and was identified as a cost-effective solution despite minor overfitting.

Conclusions:

• Meta-heuristic algorithms like TLBO and SPBO are effective for optimizing DNNs in EEG-based driver drowsiness detection.
• SPBO offers an efficient and cost-effective approach for optimizing deep learning models for road safety applications.
• The findings advance driver monitoring systems and highlight the utility of meta-heuristic techniques in deep learning.