Table 1. The experimental results (mean) on human protein data sets for investigating the usefulness of the correlations among the locations.
Evaluation metric | The proposed algorithm | ||||||
The original data set | The new data set (40%) | ||||||
Normal | Variation | The gap | Normal | Variation | The gap | ||
The whole test set | Average precision | 0.661 | 0.655 | 0.006 | 0.653 | 0.636 | 0.017 |
Recall | 0.595 | 0.587 | 0.008 | 0.562 | 0.543 | 0.019 | |
F1-score | 0.530 | 0.522 | 0.008 | 0.516 | 0.504 | 0.012 | |
Absolute true success rate | 0.274 | 0.261 | 0.013 | 0.204 | 0.189 | 0.015 | |
Coverage | 2.003 | 2.047 | −0.044 | 2.630 | 2.711 | −0.081 | |
Ranking loss | 0.129 | 0.132 | −0.003 | 0.143 | 0.148 | −0.005 | |
Samples withmultiple sites | Average precision | 0.688 | 0.673 | 0.015 | 0.700 | 0.678 | 0.022 |
Recall | 0.478 | 0.459 | 0.019 | 0.535 | 0.498 | 0.037 | |
F1-score | 0.535 | 0.518 | 0.017 | 0.572 | 0.545 | 0.027 | |
Absolute true success rate | 0.179 | 0.148 | 0.031 | 0.231 | 0.181 | 0.050 | |
Coverage | 3.889 | 4.030 | −0.141 | 3.825 | 3.954 | −0.129 | |
Ranking loss | 0.152 | 0.158 | −0.006 | 0.148 | 0.155 | −0.007 |