EEG-based Emotion Detection Using Unsupervised Transfer Learning.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion classification via electroencephalography (EEG) can aid social integration for patients with neurological disorders like ALS and Alzheimer's.
• Challenges in EEG-based emotion recognition include low signal-to-noise ratio (SNR), limited data duration, and high subject-to-subject variability.
• Existing subject-independent emotion classifiers have limitations in accuracy.

Purpose of the Study:

• To present a novel, integrated framework for semi-generic emotion detection using EEG signals.
• To address the challenges of SNR, data duration, and subject variability in EEG emotion recognition.
• To improve the accuracy of emotion recognition for individuals with neurological conditions.

Main Methods:

• EEG signal preprocessing using independent component analysis (ICA).
• EEG subject clustering via unsupervised learning techniques.
• EEG-based emotion recognition utilizing a convolutional neural network (CNN) with transfer learning.
• Data integration from DEAP, DREAMER repositories, and a local IAPS dataset.

Main Results:

• The proposed CNN classifier achieved an average accuracy of 70.26% for valence and 72.42% for arousal.
• The framework demonstrated superior performance compared to existing generic (subject-independent) emotion classifiers.
• The integrated approach effectively handled EEG data challenges, including SNR and variability.

Conclusions:

• The developed framework offers a promising advancement in EEG-based emotion detection, particularly for clinical applications.
• Semi-generic emotion recognition using this integrated approach can enhance the quality of life for patients with neurological disorders.
• The study highlights the potential of combining advanced signal processing, unsupervised learning, and deep learning for robust emotion classification.