EEG-based emotion estimation using Bayesian weighted-log-posterior function and perceptron convergence algorithm

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion recognition from electroencephalogram (EEG) signals is a challenging area.
• Emotions are commonly represented using valence and arousal dimensions.
• Accurate emotion classification requires robust feature extraction and classification methods.

Purpose of the Study:

• To propose a novel methodology for emotion recognition from EEG signals.
• To evaluate the performance of the proposed method in classifying emotions on valence and arousal dimensions.
• To compare classification accuracy for two-level and three-level emotion classes.

Main Methods:

• Feature extraction using Fast Fourier Transform (FFT).
• Feature selection utilizing Pearson correlation coefficient.
• Development of a probabilistic classifier based on Bayes' theorem.
• Supervised learning employing a perceptron convergence algorithm.
• Validation using an open EEG database.

Main Results:

• For two-level classification, average accuracy for valence estimation was 70.9% and for arousal estimation was 70.1%.
• For three-level classification, average accuracy for valence estimation was 55.4% and for arousal estimation was 55.2%.
• The proposed methodology demonstrates effective emotion recognition capabilities.

Conclusions:

• The proposed method, combining FFT feature extraction and a Bayes classifier, shows promise for EEG-based emotion recognition.
• Classification accuracy is higher for simpler, two-level emotion categorizations compared to more granular, three-level categorizations.
• The study validates the methodology using an open database, contributing to the field of affective computing.