Artificial intelligence based BCI using SSVEP signals with single channel EEG.

Background: Brain-Computer Interfaces (BCIs) enable direct communication between the brain and external devices. Steady-state visual-evoked potentials (SSVEPs) are particularly useful in BCIs because of their rapid communication capabilities and minimal calibration requirements. Although SSVEP-based BCIs are highly effective, traditional classification methods face challenges in maintaining high accuracy with minimal EEG channels, especially in real-world applications. There is a growing need for improved classification techniques to enhance performance and efficiency.

Objective: The aim of this research is to improve the classification of SSVEP signals using machine-learning algorithms. This involves extracting dominant frequency features from SSVEP data and applying classifiers such as Decision Tree (DT), Linear Discriminant Analysis (LDA), and Support Vector Machine (SVM) to achieve high accuracy while reducing the number of EEG channels required, making the method practical for BCI applications.

Methods: SSVEP data were collected from the Benchmark Dataset at Tsinghua BCI Lab using 64 EEG channels per subject. The Oz channel was selected as the dominant channel for analysis. Wavelet decomposition (db4) was used to extract frequency features in the range 7.8 Hz to 15.6 Hz. The frequency of the maximum amplitude within a 5-s window was extracted as the key feature, and machine learning models (DT, LDA, and SVM) were applied to classify these features.

Results: The proposed method achieved a high classification accuracy, with 95.8% for DT and 96.7% for both LDA and SVM. These results show significant improvement over existing methods, indicating the potential of this approach for BCI applications.

Conclusion: This study demonstrates that SSVEP classification using machine-learning models improves accuracy and efficiency. The use of wavelet decomposition for feature extraction and machine learning for classification offers a robust method for SSVEP-based BCIs. This method is promising for assistive technologies and other BCI applications.