| Network dataset |
MeGaFiller |
Non-weighted |
Common-2 |
| iIN800 |
88.27 |
86.71 |
86.40 |
| iWV1314 |
62.67 |
60.48 |
59.86 |
| iHD666 |
61.55 |
58.77 |
58.39 |
| iMA871 |
58.86 |
54.53 |
53.64 |
| iIB711 |
67.91 |
67.14 |
66.91 |
|
| F2 score [%] of our weighted ensemble (MeGaFiller) is always better than that of non-weighted version and Common-2 voted version. The non-weighted version was run the same as weighted version, except that the weights were fixed equally. The common voted version was run by taking only predictions that were made by at least 2 component gap fillers (ignoring both weights and scores). The largest value for each row is shown in bold. The non-weighted version slightly performed better than the Common-2 voted version. |
| Table 4: Performance of different variants of our ensemble gap fillers. |
