Enhancing EEG-Based Imagined Speech Recognition Through Spatio-Temporal Feature Extraction Using Information Set Theory
Abstract
We present a novel approach to imagined speech classification using EEG signals by leveraging advanced spatio-temporal feature extraction through Information Set Theory techniques. Our method enhances feature extraction and selection, significantly improving classification accuracy while reducing dataset size. Tested on the KaraOne database, our approach achieved average accuracies of 60%−90% across five phonological tasks, with the Random Forest Classifier performing best. This improvement over baseline methods illustrates the effectiveness of our feature selection techniques.
Keywords
- Imagined speech classification
- EEG signals
- Spatio-temporal features
- Information set theory
- Hanman classifier
- Machine learning