Table 2. .
Classification accuracy for each pair of training and testing sets. The training sets were composed of 90% of the samples coming from a specific network (version 1, 2, or 3), or 90% of all samples (all versions). The testing sets contained the remaining samples from the network version that are not used for training, that is, the remaining 10%. The number of samples in every testing set is reported in parentheses.
| Training set | Testing set | |||
|---|---|---|---|---|
| Ver. 1 | Ver. 2 | Ver. 3 | All ver. | |
| Version 1 | 100% (28) | 86.67% (180) | 91.18% (170) | 89.68% (378) |
| Version 2 | 97.69% (130) | 100% (24) | 93.33% (240) | 95.18% (394) |
| Version 3 | 99.23% (130) | 99.17% (240) | 100% (24) | 99.23% (394) |
| All versions | 100% (28) | 100% (24) | 100% (24) | 100% (76) |