Table 1 Comparison of the different methods in terms of mean ACC and NMI over 20 runs of a k-means clustering for several subsets with a different number d of selected features
From: Improvement of variables interpretability in kernel PCA
lapl | NDFS | KPCA-permute | KPCA-IG | |
|---|---|---|---|---|
Glioma (\(n=50, p =4434)\) | ||||
ACC(10) | \(\varvec{0.50}\) (0.02) | 0.37 (0.04) | 0.48 (0.03) | 0.42 (0.01) |
NMI(10) | \(\varvec{0.34}\) (0.02) | 0.13 (0.03) | 0.31 (0.02) | 0.21 (0.01) |
ACC(150) | \(\varvec{0.56}\) (0.03) | 0.53 (0.04) | 0.54 (0.04) | \(\varvec{0.56 (0.05)}\) |
NMI(150) | \(\varvec{0.50}\) (0.02) | 0.41 (0.03) | 0.48 (0.02) | 0.36 (0.05) |
ACC(300) | 0.54 (0.04) | 0.55 (0.04) | 0.52 (0.03) | \(\varvec{0.57}\) (0.05) |
NMI(300) | \(\varvec{0.48}\) (0.03) | 0.41 (0.03) | 0.45 (0.02) | 0.35 (0.05) |
CPU time | 0.4 | 84.6 | 620.9 | 2.9 |
Carcinom (\(n=174, p =9182)\) | ||||
ACC(10) | 0.27 (0.02) | 0.47 (0.04) | 0.48 (0.02) | \(\varvec{0.51}\) (0.01) |
NMI(10) | 0.23 (0.01) | 0.48 (0.03) | 0.43 (0.01) | \(\varvec{0.49}\) (0.02) |
ACC(150) | 0.61 (0.03) | 0.68 (0.04) | 0.67 (0.03) | \(\varvec{0.70}\) (0.02) |
NMI(150) | 0.62 (0.03) | \(\varvec{0.72}\) (0.03) | 0.69 (0.03) | 0.70 (0.03) |
ACC(300) | 0.69 (0.04) | 0.69 (0.04) | \(\varvec{0.70}\) (0.048) | 0.69 (0.03) |
NMI(300) | 0.73 (0.03) | \(\varvec{0.73}\) (0.03) | 0.71 (0.03) | 0.70 (0.02) |
CPU time | 1.4 | 391.8 | 7937.6 | 30.5 |
GPL93 (\(n=165, p =12626)\) | ||||
ACC(10) | 0.38 (0.01) | 0.41 (0.01) | \(\varvec{0.42}\) (0.01) | 0.40 (0.01) |
NMI(10) | 0.08 (0.01) | 0.109 (0.07) | 0.11 (0.01) | \(\varvec{0.15}\) (0.01) |
ACC(150) | 0.38 (0.01) | 0.45 (0.01) | \(\varvec{0.60}\) (0.02) | 0.58 (0.01) |
NMI(150) | 0.07 (0.01) | 0.18 (0.02) | \(\varvec{0.39}\) (0.01) | 0.29 (0.01) |
ACC(300) | 0.37 (0.01) | 0.49 (0.03) | 0.56 (0.05) | \(\varvec{0.56}\) (0.01) |
NMI(300) | 0.07 (0.01) | 0.22 (0.02) | 0.22 (0.01) | \(\varvec{0.31}\) (0.01) |
CPU time | 2.1 | 1277.4 | 17691.4 | 39.8 |