Skip to main content

Advertisement

Table 2 Cross validated (32 runs of 3:2 train-test split) linear SVM and random forest performance on original datasets, RST transformed data sets, LDA transformed data sets and PCA transformed data sets

From: A subspace recursive and selective feature transformation method for classification tasks

   Original rst lda pca
  No. features Log loss ± stdev Log loss ± stdev Log loss ± stdev Log loss ± stdev
Sonar
 RandomForest 60 (original) 0.49 ± 0.04    
  51   1.17 ± 0.38   0.77 ± 0.18
  43   4.83 ± 1.42   0.60 ± 0.03
  36   6.17 ± 1.31   0.57 ± 0.02
  30   6.18 ± 1.98   0.65 ± 0.15
  25   5.78 ± 1.54   0.64 ± 0.15
  20   6.03 ± 1.16   0.78 ± 0.37
  1    9.23 ± 1.21  
 LinearSVM 60 (original) 0.52 ± 0.04    
  51   0.56 ± 0.06   0.55 ± 0.04
  43   0.82 ± 0.22   0.55 ± 0.04
  36   0.85 ± 0.21   0.57 ± 0.02
  30   0.85 ± 0.20   0.55 ± 0.04
  25   0.85 ± 0.20   0.55 ± 0.04
  20   0.85 ± 0.20   0.56 ± 0.03
  1    1.00 ± 0.18  
Digits
 RandomForest 64 (original) 0.46 ± 0.07    
  54   1.00 ± 0.22   0.82 ± 0.07
  46   0.92 ± 0.36   0.81 ± 0.17
  39   0.98 ± 0.20   0.73 ± 0.08
  33   0.93 ± 0.19   0.74 ± 0.04
  27   0.87 ± 0.15   0.61 ± 0.08
  22   0.97 ± 0.20   0.59 ± 0.05
  9    0.61 ± 0.04  
 LinearSVM 64 (original) 0.25 ± 0.02    
  54   0.29 ± 0.01   0.36 ± 0.02
  46   0.24 ± 0.02   0.38 ± 0.01
  39   0.23 ± 0.02   0.36 ± 0.03
  33   0.24 ± 0.02   0.39 ± 0.03
  27   0.23 ± 0.02   0.39 ± 0.01
  22   0.24 ± 0.02   0.41 ± 0.02
  9    0.28 ± 0.02  
Letter
 RandomForest 16 (original) 0.64 ± 0.01    
  12   2.78 ± 0.11 1.36 ± 0.05 1.05 ± 0.02
  9   7.17 ± 0.05 1.61 ± 0.05 1.20 ± 0.03
  5   7.08 ± 0.05 2.28 ± 0.07 1.96 ± 0.06
  2   7.12 ± 0.16 4.11 ± 0.07 4.98 ± 0.09
 LinearSVM 16 (original) 1.33 ± 0.01    
  12   1.33 ± 0.01 1.62 ± 0.02 1.45 ± 0.01
  9   1.37 ± 0.01 1.89 ± 0.02 1.66 ± 0.01
  5   1.33 ± 0.03 2.20 ± 0.04 2.05 ± 0.02
  2   1.28 ± 0.01 2.51 ± 0.01 2.44 ± 0.01
Glass
 RandomForest 9 (original) 2.43 ± 0.83    
  5   4.33 ± 1.61 3.53 ± 0.83 2.44 ± 0.88
  2   5.80 ± 0.45 3.82 ± 0.61 3.18 ± 1.15
 LinearSVM 9 (original) 1.35 ± 0.04    
  5   1.23 ± 0.04 1.20 ± 0.03 1.18 ± 0.03
  2   1.10 ± 0.04 1.22 ± 0.03 1.21 ± 0.05
Iris      
 RandomForest 4 (original) 0.65 ± 0.36    
  2   0.39 ± 0.68 0.25 ± 0.22 0.28 ± 0.27
 LinearSVM 4 (original) 0.46 ± 0.04    
  2   0.30 ± 0.04 0.43 ± 0.02 0.48 ± 0.05
  1. In bold are the best score (lowest log loss) on the corresponding data set. Results for sonar: rock vs mine sensory readings, hand written digits, hand written letter recognition, glass identification and iris