Table 3.
The 2D raw EEG signal for 2D CNN models results. In contrast to the image-based input, a slight increase in performance can be observed in accuracy. Most of the devised representations were able to achieve performances higher than 50%. The best results obtained with this form of input representation were obtained with the 2D raw 9- and 5-channel EEG signal sample, with an accuracy of 62.22% and 63.31% in combination with the single-layer CNN model. The three-layer CNN model with the 9- and 5-channel selections was able to reach a classification accuracy of 63.28% and 61.26% respectively. In the case of the two-layer CNN model in combination with the 64-channel selection, it was able to return a classification accuracy of 59.68%. Lastly, the only one that was able to achieve a result that outperformed the state of the art was that of the two-layer CNN model in combination with the 9-channel selection of the 2D raw EEG signal, with a classification accuracy of 72.87%. Results that outperform the state of the art are highlighted in bold.
| Model | Input | Accuracy [%] |
|---|---|---|
| [25] | Topographical Map | 65.00% |
| [26] | 2D Raw EEG signal | 69.82% |
| Single-Layer CNN | 2D Raw 64-Channel EEG Signal | 55.23% |
| 2D Raw 21-Channel EEG Signal | 53.54% | |
| 2D Raw 13-Channel EEG Signal | 53.75% | |
| 2D Raw 9-Channel EEG Signal | 62.22% | |
| 2D Raw 5-Channel EEG Signal | 63.31% | |
| Two-Layer CNN | 2D Raw 64-Channel EEG Signal | 59.68% |
| 2D Raw 21-Channel EEG Signal | 55.66% | |
| 2D Raw 13-Channel EEG Signal | 52.16% | |
| 2D Raw 9-Channel EEG Signal | 72.87% | |
| 2D Raw 5-Channel EEG Signal | 54.39% | |
| Three-Layer CNN | 2D Raw 64-Channel EEG Signal | 60.42% |
| 2D Raw 21-Channel EEG Signal | 53.12% | |
| 2D Raw 13-Channel EEG Signal | 54.25% | |
| 2D Raw 9-Channel EEG Signal | 63.28% | |
| 2D Raw 5-Channel EEG Signal | 61.26% |