Table 3.
Baseline results on various machine learning and deep learning models for identification and verification.
| Identification | ||||||
|---|---|---|---|---|---|---|
| Models | EEG | Wacom | EEG + Wacom | |||
| Statistical features | Wavelet features | Wacom eatures | Statistical + Wacom | Wavelet + Wacom | Statistical+Wavelet Wacom | |
| Gaussian NB | 93.4 ± 0 | 90.24 ± 0 | 21.90 ± 0 | 96.31 ± 0 | 96.51 ± 0 | 96.57 ± 0 |
| KNN | 91.82 ± 0 | 51.98 ± 0 | 25.33 ± 0 | 93.14 ± 0 | 51.98 ± 0 | 59.10 ± 0 |
| Linear SVM | 91.82 ± 0 | 89.71 ± 0 | 10.82 ± 0 | 96.83 ± 0 | 91.29 ± 0 | 93.93 ± 0 |
| Random Forest | 91.66 ± 0.0043 | 91.08 ± 0.0045 | 55.25 ± 0.0125 | 97.63 ± 0.006 | 97.68 ± 0.0019 | 98.94 ± 0.0033 |
| Gradient Boost | 92.88 ± 0.0013 | 90.82 ± 0.002 | 38.58 ± 0.0112 | 96.73 ± 0.0013 | 94.78 ± 0.0011 | 94.35 ± 0.0027 |
| XGBoost | 93.93 ± 0 | 92.08 ± 0 | 51.19 ± 0 | 98.68 ± 0 | 97.89 ± 0 | 97.89 ± 0 |
| Verification | ||||||
| Gaussian NB | 67.28 ± 0 | 63.14 ± 0 | 92.78 ± 0 | 92.52 ± 0 | 92.52 ± 0 | 91.49 ± 0 |
| KNN | 68.04 ± 0 | 68.04 ± 0 | 88.91 ± 0 | 74.48 ± 0 | 68.04 ± 0 | 68.29 ± 0 |
| Linear SVM | 73.40 ± 0 | 60.30 ± 0 | 84.02 ± 0 | 86.64 ± 0 | 68.04 ± 0 | 70.10 ± 0 |
| Random Forest | 75.30 ± 0.0078 | 69.38 ± 0.0034 | 95.05 ± 0.0022 | 94.84 ± 0.0036 | 94.69 ± 0.0029 | 94.89 ± 0.0056 |
| Gradient Boost | 68.50 ± 0.0067 | 65.30 ± 0.005 | 93.81 ± 0.0026 | 93.55 ± 0.0045 | 93.50 ± 0.0042 | 93.40 ± 0.0029 |
| XGBoost | 68.04 ± 0 | 63.14 ± 0 | 91.49 ± 0 | 92.01 ± 0 | 91.49 ± 0 | 92.26 ± 0 |
The accuracy score in mean and standard deviation for five trials of each experiment are shown. This is a performance analysis on the data of Task-4, i.e., EEG collected while physically drawing the signature on the Wacom tablet.