Table 4. Classification errors using (a) global features and (b) activation and degree maps (using SVM on the complete set of voxels (i.e., without voxel subset selection).
| (a) | |||
| Feature | GNB | SVM | MRF(0.01) |
| degree (D) | 27.5% | 27.5% | 27.5% |
| clustering coeff. (C) | 30.0% | 42.5% | 45.0% |
| geodesic dist. (G) | 67.5% | 45.0% | 45.0% |
| mean activation (A) | 40.0% | 45% | 72.5% |
| D+A | 27.5% | 27.5% | 32.5% |
| C+A | 27.5% | 45.0% | 55.0% |
| G+A | 45.0% | 45.0% | 72.5% |
| G +D +C | 37.5% | 27.5% | 27.5% |
| G+D+C+A | 30.0% | 27.5% | 32.5% |
| (b) | |||
| Feature | Err | FP | FN |
| correlations (53750) | 14% | 14% | 14% |
| degree (full) | 16% | 27% | 5% |
| degree (long-distance) | 21% | 32% | 9% |
| degree (inter-hemis) | 32% | 46% | 18% |
| clustering | 23% | 32% | 14% |
| local efficiency | 23% | 32% | 14% |
| strength | 23% | 23% | 23% |
| abs strength | 34% | 41% | 27% |
| pos strength | 25% | 32% | 18% |
| activation 1 (and 3) | 54% | 29% | 82% |
| activation 2 (and 4) | 50% | 55% | 45% |
| activation 5 | 43% | 18% | 68% |
| activation 6 | 36% | 27% | 46% |
| activation 7 | 32% | 18% | 46% |
| activation 8 | 30% | 23% | 37% |
For each feature, we show the average error, as well as the fraction of false positives (FP) and false negatives (FN).