Ranking
Based on a thorough evaluation using 104 imbalanced datasets, the following 10 techniques provide the highest performance in terms of the AUC, GAcc, F1 and P20 scores, in nearest neighbors, support vector machine, decision tree and multilayer perceptron based classification scenarios. For more details on the evaluation methodology, see our paper on the comparative study.
sampler |
overall |
auc |
rank_auc |
gacc |
rank_gacc |
f1 |
rank_f1 |
p_top20 |
rank_ptop20 |
|---|---|---|---|---|---|---|---|---|---|
polynom-fit-SMOTE |
2.5 |
0.902538 |
6 |
0.870753 |
1 |
0.695154 |
1 |
0.992496 |
2 |
ProWSyn |
4.5 |
0.904389 |
1 |
0.868449 |
4 |
0.690284 |
3 |
0.991112 |
10 |
SMOTE-IPF |
7.5 |
0.902565 |
5 |
0.868715 |
3 |
0.687935 |
9 |
0.990904 |
13 |
Lee |
8 |
0.902318 |
7 |
0.868324 |
5 |
0.688082 |
8 |
0.991008 |
12 |
SMOBD |
9.25 |
0.902247 |
8 |
0.86766 |
6 |
0.688885 |
4 |
0.990583 |
19 |
G-SMOTE |
13.5 |
0.901916 |
10 |
0.865103 |
18 |
0.686613 |
12 |
0.990846 |
14 |
CCR |
14.25 |
0.902112 |
9 |
0.861994 |
30 |
0.687886 |
10 |
0.991254 |
8 |
LVQ-SMOTE |
14.75 |
0.902799 |
3 |
0.862295 |
29 |
0.683646 |
24 |
0.992211 |
3 |
Assembled-SMOTE |
15.5 |
0.902691 |
4 |
0.866914 |
7 |
0.688614 |
5 |
0.982685 |
46 |
SMOTE-TomekLinks |
15.75 |
0.901016 |
14 |
0.866174 |
9 |
0.684708 |
20 |
0.990573 |
20 |