Skip to main content

Advertisement

Table 1 Average classification accuracy and standard deviation.

From: ANMM4CBR: a case-based reasoning method for gene expression data classification

# Iteration 10 20 30 40 50
Leukemia BW+k NN 95.7 ± 1.2 96.9 ± 1.8 96.6 ± 2.2 96.6 ± 1.2 96.8 ± 1.7
  MRMR+k NN 96.5 ± 2.5 96.4 ± 2.1 97.4 ± 1.7 96.9 ± 2.2 95.8 ± 2.4
  BW+SVM 95.6 ± 1.3 95.7 ± 1.7 95.9 ± 2.2 96.2 ± 2.3 96.9 ± 1.2
  MRMR+SVM 96.4 ± 2.5 96.8 ± 3.6 97.6 ± 2.0 97.1 ± 2.7 96.8 ± 3.4
  LogitBoost 95.3 ± 2.9 96.0 ± 2.4 96.6 ± 1.8 96.6 ± 2.8 96.7 ± 1.7
  ANMM4CBR 96.3 ± 2.4 97.5 ± 1.7 97.3 ± 1.8 96.6 ± 1.7 97.0 ± 2.3
Colon BW+k NN 81.2 ± 8.1 82.8 ± 7.5 83.5 ± 4.2 83.4 ± 5.3 83.6 ± 6.5
  MRMR+k NN 83.7 ± 4.3 83.6 ± 7.9 84.2 ± 6.0 83.8 ± 5.9 83.5 ± 6.9
  BW+SVM 84.0 ± 4.3 83.6 ± 6.4 83.6 ± 6.0 84.2 ± 7.2 84.5 ± 7.9
  MRMR+SVM 85.4 ± 5.8 84.1 ± 6.6 84.0 ± 4.0 84.6 ± 7.0 84.7 ± 8.1
  LogitBoost 84.4 ± 4.3 84.5 ± 8.9 83.6 ± 4.9 84.2 ± 6.8 84.1 ± 4.6
  ANMM4CBR 86.3 ± 6.1 86.7 ± 5.6 86.2 ± 4.2 86.5 ± 5.6 85.6 ± 4.4
SRBCT BW+k NN (50) 94.4 ± 4.2 97.7 ± 2.1 97.9 ± 1.3 98.2 ± 1.6 98.0 ± 1.2
  MRMR+k NN (50) 78.4 ± 9.0 97.4 ± 1.9 98.6 ± 1.0 98.8 ± 0.9 98.2 ± 0.8
  BW+SVM (97) 94.0 ± 3.2 98.0 ± 1.4 98.4 ± 1.2 98.8 ± 0.9 99.2 ± 0.3
  MRMR+SVM (95) 81.0 ± 10.5 98.2 ± 1.0 98.9 ± 1.3 99.1 ± 0.7 99.2 ± 0.2
  LogitBoost (102) 94.9 ± 3.1 97.3 ± 1.8 98.0 ± 1.6 98.6 ± 1.1 98.6 ± 0.6
  ANMM4CBR (50) 90.3 ± 5.5 97.3 ± 1.5 98.8 ± 1.2 99.3 ± 0.7 99.7 ± 0.3
GCM BW+k NN (50) 46.2 ± 4.7 47.4 ± 7.0 51.2 ± 4.9 52.6 ± 6.2 54.1 ± 5.8
  MRMR+k NN (50) 41.1 ± 7.1 42.7 ± 8.1 51.5 ± 1.6 58.3 ± 4.9 60.5 ± 5.9
  BW+SVM (254) 53.7 ± 5.1 58.1 ± 9.8 59.0 ± 6.6 66.6 ± 6.7 66.9 ± 3.6
  MRMR+SVM (259) 51.0 ± 7.7 60.3 ± 7.0 61.8 ± 3.7 64.8 ± 8.2 67.8 ± 4.6
  LogitBoost (273) 57.1 ± 4.9 60.1 ± 1.9 60.6 ± 4.0 62.1 ± 5.7 65.1 ± 5.4
  ANMM4CBR (50) 41.1 ± 1.2 51.0 ± 8.1 57.2 ± 6.9 61.1 ± 1.4 63.3 ± 3.9
  1. Each experiment was carried out for 100 runs. The best results in different situations are labeled as black. Here the iteration number means the number of features used by each single classifier. In OVA case, the total number of genes may exceed the iteration number, since in OVA a multiclass problem is solved by considering many binary ones. In the parentheses we list the average number of features selected by each method when the iteration number is 50. See Table 2 for another experiment on multiclass data set.