Multi-Input CNN-LSTM deep learning model for fear level classification based on EEG and peripheral physiological...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Accurate fear level classification is crucial for treating anxiety disorders like PTSD and phobias.
• Current methods may lack objectivity or require manual feature engineering.
• Developing automated, high-accuracy fear detection is a significant clinical need.

Purpose of the Study:

• To develop and validate a deep learning model for automatic, high-accuracy fear level estimation.
• To investigate the efficacy of a novel Multi-Input Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture.
• To assess the model's performance without manual feature extraction and its adaptability to individual differences.

Main Methods:

• Utilized the DEAP dataset comprising multichannel electroencephalography (EEG) and peripheral physiological signals.
• Implemented a Multi-Input CNN-LSTM deep learning model for classification.
• Performed 10-fold cross-validation to evaluate model performance.

Main Results:

• The Multi-Input CNN-LSTM model achieved 98.79% accuracy and a 99.01% F1 score in classifying four distinct fear levels.
• Demonstrated the feasibility of high-accuracy emotion recognition directly from raw physiological signals.
• Showcased the model's potential for generalization across individuals, with possibilities for further accuracy enhancement.

Conclusions:

• Deep learning, specifically the Multi-Input CNN-LSTM model, offers a highly accurate method for automated fear level recognition from physiological data.
• This approach bypasses the need for subjective feature selection, enabling objective emotion assessment.
• The model shows promise for clinical applications in diagnosing and managing fear-related and anxiety disorders.