A Deep-Learning Model for Subject-Independent Human Emotion Recognition Using Electrodermal Activity Sensors

Area of Science:

• Physiology and Computer Science
• Focuses on affective computing and biosignal processing.

Background:

• Active and Assisted Living (AAL) environments aim to support elderly and disabled individuals.
• Emotion recognition using non-intrusive sensors like Electrodermal Activity (EDA) is crucial for monitoring well-being.
• Subject-independent emotion recognition remains a significant challenge due to variations in individual emotional patterns.

Purpose of the Study:

• To propose a robust Convolutional Neural Network (CNN) architecture for human emotion recognition.
• To achieve reliable performance in both subject-dependent and subject-independent scenarios.
• To validate the effectiveness of using solely EDA sensors for emotion classification.

Main Methods:

• Developed a novel CNN architecture for emotion recognition.
• Employed a grid search technique for hyperparameter optimization of the CNN model.
• Validated the model's performance using the MAHNOB and DEAP datasets.

Main Results:

• Achieved promising robustness improvements in emotion classification accuracy.
• Reached 78% and 82% accuracy for subject-independent classification on MAHNOB and DEAP datasets, respectively.
• Attained 81% and 85% accuracy for subject-dependent classification on MAHNOB and DEAP datasets, respectively (4 classes).

Conclusions:

• Solely using non-intrusive EDA sensors enables robust human emotion classification.
• The proposed CNN architecture demonstrates significant potential for AAL applications.
• The findings highlight the feasibility of reliable emotion recognition without additional physiological signals.