Skip to main content

Advertisement

Table 4 The minimum error rates yielded by Random Forest classifier with feature selection methods along‐with the classification error without selection

From: A feature selection method for classification within functional genomics experiments based on the proportional overlapping score

Dataset ISIS Wil‐RS mRMR MP POS Full set
Leukaemia 0.003 (1) 0.030 (20) 0.118 (40) 0.015 (9) 0.0002 (40) 0.049
Breast 0.407 (4) 0.371 (50) 0.407 (48) 0.354 (48) 0.308 (45) 0.369
Srbct 0.092 (2.63) 0.069 (24) 0.074 (46) 0.009 (32) 0.003 (48) 0.0008
Prostate 0.097 (4.18) 0.200 (50) 0.140 (50) 0.069 (50) 0.062 (50) 0.088
All 0.0004 (1.018) 0.143 (40) 0.011 (50) 0 (40) 0 (20) 0
Lung 0.022 (3.26) 0.040 (30) 0.016 (48) 0.008 (46) 0.007 (48) 0.003
Carcinoma 0.171 (1.29) 0.003 (41) 0.017 (44) 0.019 (5) 0.026 (20) 0.027
GSE24514 0.107 (1.96) 0.054 (47) 0.063 (50) 0.036 (48) 0.032 (24) 0.041
GSE4045 0.27 (1.47) 0.134 (24) 0.187 (37) 0.137 (21) 0.114 (27) 0.205
GSE14333 0.423 (9) 0.421 (10) 0.438 (31) 0.437 (34) 0.414
GSE27854 0.448 (5) 0.401 (15) 0.444 (49) 0.451 (6) 0.488
  1. The numbers in brackets represent the size, average size for ISIS method, of the gene set that corresponding to the minimum error rate. Boldface numbers indicate the lowest error rate (the highest accuracy) among the compared methods for the corresponding datasets.