Table 2. G-mean measured using a hybrid learning strategy (FS + WeightMI or FS + MinCostMI) or feature selection alone (FS + MI), for (a) the DLBCL dataset, (b) the Glioma dataset and (c) the Uterus dataset.
The comparison is carried out, separately for the FS + WeightMI and FS + MinCostMI strategies, using CORR and SVM-RFE as selection methods (by retaining 0.25% and 2% of the original features). The values in bold turned out to be significantly different from those obtained with the FS + MI approach according to the Wilcoxon signed-rank test with Holm-Bonferroni correction (p-values in brackets).
| (a) | FS+ WeightMI(c) | FS+MinCostMI(c) | |||||
|---|---|---|---|---|---|---|---|
| FS+MI | c = 2 | c = 3 | c = 4 | c = 2 | c = 3 | c = 4 | |
| CORR 0.25% | 0.79 | 0.86 (0.0119) | 0.87 (0.0053) | 0.89 (0.0010) | 0.88 (0.0029) | 0.87 (0.0059) | 0.85 (0.0243) |
| CORR 2% | 0.82 | 0.86 (0.1427) | 0.86 (0.1427) | 0.90 (0.0021) | 0.89 (0.0048) | 0.90 (0.0093) | 0.87 (0.0515) |
| SVM-RFE 0.25% | 0.90 | 0.92 (0.2204) | 0.92 (0.0774) | 0.93 (0.1105) | 0.93 (0.0659) | 0.89 (0.2797) | 0.86 (0.0779) |
| SVM-RFE 2% | 0.84 | 0.91 (0.0091) | 0.93 (0.0121) | 0.96 (0.0024) | 0.96 (0.0007) | 0.94 (0.0098) | 0.89 (0.1436) |
| (b) | FS+ WeightMI(c) | FS+MinCostM/(c) | |||||
| FS+MI | c = 3 | c = 4 | c = 5 | c = 3 | c = 4 | c = 5 | |
| CORR 0.25% | 0.56 | 0.56 (1.00) | 0.64 (0.2919) | 0.66 (0.1404) | 0.67 (0.1099) | 0.75 (0.0324) | 0.86 (0.0141) |
| CORR 2% | 0.59 | 0.67 (0.1714) | 0.61 (0.3759) | 0.68 (0.0700) | 0.83 (0.0178) | 0.87 (0.0143) | 0.88 (0.0113) |
| SVM-RFE 0.25% | 0.42 | 0.56 (0.0284) | 0.53 (0.1030) | 0.55 (0.1342) | 0.73 (0.0041) | 0.76 (0.0024) | 0.78 (0.0017) |
| SVM-RFE 2% | 0.37 | 0.62 (0.0157) | 0.65 (0.0100) | 0.65 (0.0062) | 0.78 (0.0019) | 0.86 (0.0009) | 0.89 (0.0008) |
| (c) | FS+ WeightMI(c) | FS+MinCostMI(c) | |||||
| FS+MI | c = 3 | c = 4 | c = 5 | c = 3 | c = 4 | c = 5 | |
| CORR 0.25% | 0.65 | 0.74 (<0.0001) | 0.76 (<0.0001) | 0.77 (<0.0001) | 0.85 (<0.0001) | 0.86 (<0.0001) | 0.87 (<0.0001) |
| CORR 2% | 0.63 | 0.72 (<0.0001) | 0.73 (<0.0001) | 0.75 (<0.0001) | 0.84 (<0.0001) | 0.87 (<0.0001) | 0.88 (<0.0001) |
| SVM-RFE 0.25% | 0.56 | 0.67 (<0.0001) | 0.66 (<0.0002) | 0.69 (<0.0001) | 0.80 (<0.0001) | 0.84 (<0.0001) | 0.86 (<0.0001) |
| SVM-RFE 2% | 0.47 | 0.63 (<0.0001) | 0.64 (<0.0001) | 0.67 (<0.0001) | 0.80 (<0.0001) | 0.86 (<0.0001) | 0.88 (<0.0001) |