A Novel Optimized Hybrid Deep Learning Framework for Mental Stress Detection Using Electroencephalography.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Mental stress, triggered by various pressures, significantly impacts human behavior and is increasingly studied using electroencephalograms (EEGs).
• Existing deep learning (DL) methods for EEG-based stress detection face challenges like complex structures, class imbalance, and signal processing difficulties.

Purpose of the Study:

• To present a novel hybrid deep learning framework, BDDNet, for enhanced mental stress detection using EEGs.
• To improve the spectral-temporal feature depiction and long-term dependency analysis of EEG signals for stress detection.

Main Methods:

• Developed BDDNet, a hybrid framework combining a deep convolutional neural network (DCNN), bidirectional long short-term memory (BiLSTM), and deep belief network (DBN).
• Utilized multiple EEG features (MEFs) for comprehensive spectral and time-domain analysis.
• Employed an improved crow search algorithm (ICSA) for efficient channel selection and an employee optimization algorithm (EOA) for hyper-parameter tuning.

Main Results:

• The BDDNet-ICSA framework demonstrated superior performance on the public DEAP dataset.
• Achieved high metrics including 97.6% recall, 97.6% precision, 97.6% F1-score, 96.9% selectivity, 96.9% negative predictive value (NPV), and 97.3% accuracy.
• Outperformed traditional stress detection techniques.

Conclusions:

• BDDNet offers a robust and effective solution for accurate mental stress detection from EEG data.
• The hybrid deep learning approach combined with optimization algorithms significantly enhances stress detection capabilities.
• This framework holds promise for advancing non-invasive mental stress monitoring and management.