Hybrid Deep Learning Approach for Stress Detection Using Decomposed EEG Signals.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Psychological stress significantly impacts physical health and daily performance.
• Early detection of stress is crucial for preventing disease progression and improving patient outcomes.
• Electroencephalography (EEG) is a key tool for measuring brain activity related to psychological states.

Purpose of the Study:

• To develop an efficient method for automatic feature extraction and stress detection from multichannel EEG recordings.
• To investigate the efficacy of a hybrid deep learning model for classifying stress levels.
• To compare the proposed model's performance against traditional deep learning techniques.

Main Methods:

• Applied Discrete Wavelet Transform (DWT) to decompose 14-channel EEG signals, addressing non-linearity and non-stationarity.
• Utilized a Convolutional Neural Network (CNN) for automatic feature extraction from decomposed EEG bands.
• Employed a hybrid model integrating BiLSTM and two GRU layers for stress level classification.

Main Results:

• The proposed DWT-based hybrid model (CNN, BiLSTM, GRU) demonstrated superior classification accuracy compared to individual traditional models (CNN, LSTM, BiLSTM, GRU, RNN).
• The hybrid approach effectively handled long-term dependencies within non-linear EEG signals.
• The model achieved higher accuracy in distinguishing stress levels from EEG data.

Conclusions:

• Hybrid deep learning models offer a promising approach for accurate and efficient psychological stress detection.
• The proposed DWT-based CNN, BiLSTM, and GRU integration is suitable for clinical applications in mental and physical health.
• This method supports early intervention and prevention strategies for stress-related conditions.