We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
The Efficacy and Utility of Lower-Dimensional Riemannian Geometry for EEG-Based Emotion Classification.
- Authors
Al-Mashhadani, Zubaidah; Bayat, Nasrin; Kadhim, Ibrahim F.; Choudhury, Renoa; Park, Joon-Hyuk
- Abstract
Electroencephalography (EEG) signals have diverse applications in brain-computer interfaces (BCIs), neurological condition diagnoses, and emotion recognition across healthcare, education, and entertainment domains. This paper presents a robust method that leverages Riemannian geometry to enhance the accuracy of EEG-based emotion classification. The proposed approach involves adaptive feature extraction using principal component analysis (PCA) in the Euclidean space to capture relevant signal characteristics and improve classification performance. Covariance matrices are derived from the extracted features and projected onto the Riemannian manifold. Emotion classification is performed using the minimum distance to Riemannian mean (MDRM) classifier. The effectiveness of the method was evaluated through experiments on four datasets, DEAP, DREAMER, MAHNOB, and SEED, demonstrating its generalizability and consistent accuracy improvement across different scenarios. The classification accuracy and robustness were compared with several state-of-the-art classification methods, which supports the validity and efficacy of using Riemannian geometry for enhancing the accuracy of EEG-based emotion classification.
- Subjects
RIEMANNIAN geometry; ELECTROENCEPHALOGRAPHY; FEATURE extraction; EMOTION recognition; BRAIN-computer interfaces; EMOTIONS; PRINCIPAL components analysis; WAKEFULNESS
- Publication
Applied Sciences (2076-3417), 2023, Vol 13, Issue 14, p8274
- ISSN
2076-3417
- Publication type
Article
- DOI
10.3390/app13148274