Table 2.
The top 10 features that impacting the GBDT classifier mostly.
| SF | NSF | Z/t value | P -value | P ′ -value | |
|---|---|---|---|---|---|
| (A) GBDT-LZC | |||||
| δ-F8 | 0.0240 ± 0.0214 | 0.0298 ± 0.0132 | −1.656 | 0.098 | 0.098 |
| θ-T3 | 0.0518 ± 0.0356 | 0.0566 ± 0.0558 | −2.010 | 0.044 | 0.060 |
| α-Cz | 0.0631 ± 0.0383 | 0.0745 ± 0.0500 | −2.472 | 0.013 | 0.032 * |
| θ-F3 | 0.0510 ± 0.0352 | 0.0561 ± 0.0526 | −2.032 | 0.042 | 0.057 |
| α-Fz | 0.0649 ± 0.0384 | 0.0757 ± 0.0582 | −2.424 | 0.015 | 0.036 * |
| θ-T6 | 0.0531 ± 0.0394 | 0.0587 ± 0.0528 | −1.880 | 0.060 | 0.068 |
| TLE | |||||
| β-T3 | 0.1119 ± 0.0742 | 0.1261 ± 0.1133 | −2.081 | 0.037 | 0.050 |
| β-Pz | 0.1100 ± 0.7230 | 0.1227 ± 0.1024 | −2.081 | 0.037 | 0.050 |
| α-T6 | 0.0624 ± 0.0334 | 0.0742 ± 0.0569 | −2.389 | 0.017 | 0.038 * |
| (B) GBDT-KC | |||||
| δ-T3 | 0.2245 ± 0.0068 | 0.2531 ± 0.2333 | −1.809 | 0.070 | 0.070 |
| θ-F7 | 0.0503 ± 0.0204 | 0.0575 ± 0.0555 | −1.904 | 0.057 | 0.065 |
| Seizure frequency before OXC | |||||
| θ-FP1 | 0.0521 ± 0.038 | 0.0573 ± 0.0566 | −1.928 | 0.054 | 0.061 |
| θ-T6 | 0.0537 ± 0.0398 | 0.0594 ± 0.0535 | −1.904 | 0.057 | 0.065 |
| θ-Fz | 0.0505 ± 0.0371 | 0.0564 ± 0.0540 | −2.105 | 0.035 | 0.044 * |
| β-T6 | 0.1116 ± 0.0728 | 0.1250 ± 0.1084 | −2.105 | 0.035 | 0.044 * |
| β-O2 | 0.1109 ± 0.0704 | 0.1244 ± 0.1148 | −2.200 | 0.028 | 0.031 * |
| Seizure circadian rhythm | |||||
| α-Pz | 0.0645 ± 0.037 | 0.0753 ± 0.0542 | −2.306 | 0.021 | 0.028 * |
GBDT, gradient boosting decision tree; LZC, Lempel-Ziv complexity; KC, Kolmogorov complexity; SF, seizure-free; NSF, not seizure-free; TLE, temporal lobe epilepsy; δ-F8, δ band from F8 channel; OXC, oxcarbazepine; P′-value refers to P-value that is corrected by false discovery rate correction.
The features that have statistically significance. Although the selected features may not be statistically significant, they did have a classification value in the model.
Bold values are statistically significant.