The Effect of Time Window Length on EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Previous electroencephalogram (EEG) studies used varied time window (TW) lengths for feature extraction.
• The impact of TW length on EEG feature extraction for emotion recognition remains underexplored.

Purpose of the Study:

• To investigate the effect of different TW lengths on human emotion recognition using EEG signals.
• To determine the optimal TW length for extracting EEG-based emotion signals.
• To evaluate the efficacy of experiment-level batch normalization (ELBN) in improving emotion recognition.

Main Methods:

• Utilized the SJTU emotion EEG dataset (SEED).
• Extracted power spectral density (PSD) and differential entropy (DE) features across various TW lengths.
• Applied experiment-level batch normalization (ELBN) to processed features.
• Performed emotion recognition using six classifiers, comparing results with and without ELBN.

Main Results:

• A 2-second TW length demonstrated the best performance for EEG emotion recognition.
• ELBN significantly improved emotion recognition accuracy by 21.63% (PSD features) and 5.04% (DE features) with a 2-s TW using SVM.
• The 2-s TW is identified as the most suitable for EEG feature extraction in emotion recognition tasks.

Conclusions:

• The optimal time window length for EEG feature extraction in emotion recognition is 2 seconds.
• Experiment-level batch normalization effectively enhances EEG-based emotion recognition performance.
• Findings offer a practical reference for selecting TW length in EEG signal analysis for intelligent systems.