Skip to main content

Table 4 Classification accuracies (%) on 9 real data sets

From: Feature weight estimation for gene selection: a local hyperlinear learning approach

Method Datasets Average
  CNS Colon DLBCL GCM Leukemia Lung Prostate1 Prostate2 Prostate3  
TSP [20] 77.90 91.10 98.10 75.40 93.80 98.30 95.10 67.60 97.00 88.26
k-TSP [19] 97.10 90.30 97.40 85.40 95.83 98.90 91.18 75.00 97.00 92.01
PAM [19] 82.35 89.52 85.45 82.32 94.03 97.90 90.89 81.25 94.24 88.66
sumdiff-PAM [21] 79.41 87.10 87.01 83.57 95.83 98.34 93.14 77.27 96.97γ 88.74
mul-PAM [21] 85.29 90.32 92.21 82.86 95.83 98.90 92.16 79.55 93.94 90.12
sign-PAM [21] 85.29 88.71 94.81 81.07 95.83 98.90 90.20 76.14 100 90.11
HBE [22]    96.10   98.61   96.08    
IVGA-SVM [23]   91.61    97.22   92.06    
BBF-SVM [24]   87.10 92.71     94.12    
SVM [25] 82.35 83.87 96.10 93.57 98.61 98.90 91.18 76.14 100 91.19
NB [25] 79.41 58.06 79.22 82.5 98.61 98.34 62.75 80.68 93.94 81.50
BMSF-SVM [25] 94.12 95.16 97.40 98.57 98.61 99.45 97.06 98.86 100 97.69
BMSF-LDA [25] 97.06 87.10 96.10 90.36 98.61 97.79 95.10 94.32 96.97η 94.82
BMSF-QDA [25] 97.06 90.32 94.81 90.36 97.22 97.23 94.12 90.91 100 94.67
BMSF-NB [25] 94.12 87.10 88.31 87.86 95.83 98.90 89.22 89.77 100 92.34
LHR-SVMξ 100 87.10 94.81 100 98.61 100 96.08 95.45 100 96.89
LHR-LDAξ 99.47 87.38 95.00 99.44 98.75 99.47 97.09 95.42 99.47 96.83
LHR-NBξ 97.79 90.32 92.21 97.24 98.61 97.24 98.04 89.77 97.79 95.44
LHR-KNNξ 98.45 91.94 96.10 91.00 100 100 99.02 94.32 99.45 96.70
LHR-HKNNξ 100 90.32 97.40 97.40 100 100 97.06 94.32 100 97.39
I-RELIEF-SVMη[12] 83.43 75.81 92.21 92.21 94.44 83.98 88.24 82.95 81.12 86.04
I-RELIEF-LDAη[12] 81.17 74.05 89.46 89.46 92.86 80.06 80.64 87.50 80.18 83.93
I-RELIEF-NBη[12] 85.08 67.74 84.42 84.42 91.67 86.74 73.53 81.82 87.29 82.52
I-RELIEF-KNNη[12] 88.4 82.26 96.10 96.10 94.44 88.40 91.18 86.36 87.85 90.12
I-RELIEF-HKNNη[12] 83.98 77.42 96.10 96.10 95.83 86.16 85.29 77.27 83.98 86.90
  1. ξClassification with our selected genes.
  2. ηClassification with selected genes by [11].
  3. γThe value of 96.97 in [26] could have been rounded to 97.00 and is suboptimal.
  4. The optimal and suboptimal values on each tested data are highlighted in bold and italic, respectively. The averaged performance of the proposed method with HKNN classifier is suboptimal to BMSM-SVM by a neglectable difference. Besides, the averaged performance of LHR, coupling with five classifiers show a dramatically smaller variance (0.725) than other BMSM does (2.191), thus implying a high capability of stability with respect to classification models.