Combining Inter-Subject Modeling with a Subject-Based Data Transformation to Improve Affect Recognition from EEG...

Area of Science:

• Neuroscience and Affective Computing

Background:

• Electroencephalography (EEG) signals correlate with emotions, leading to various affect detection methods.
• Current methods include intra-subject (user-specific) and inter-subject (generalized) approaches.
• Intra-subject methods face data limitations, while inter-subject methods neglect individual differences.

Purpose of the Study:

• To analyze the performance of intra-subject and inter-subject affect detection models.
• To investigate the influence of individual subjects on EEG signals compared to emotional content.
• To propose a novel method for integrating individual traits into inter-subject models.

Main Methods:

• Analysis of three public EEG repositories.
• Comparative study of intra-subject and inter-subject modeling techniques.
• Development of a data transformation to incorporate subject-specific features.

Main Results:

• Subject-specific influence on EEG signals is significantly greater than emotion-specific influence.
• Inter-subject models alone are insufficient due to high individual variability.
• The proposed data transformation enhances classification accuracy by accounting for individual traits.

Conclusions:

• Subject-specific modeling is crucial for accurate EEG-based affect detection.
• A data transformation method effectively integrates individual characteristics into generalized models.
• This approach improves the robustness and applicability of inter-subject affect detection systems.