How to combine Base Classifiers <3> : James-Stein vs MLE Estimator

Table 1 Comparing MUV(X, <boot,bal>, mle) vs MUV(X, <boot,bal>, js)

2NBs 3NBs 4NBs 2TANs 3TANs 4TANs
(Boot,Bal) MLE JS MLE JS MLE JS MLE JS MLE JS MLE JS
australian 86.06 87.68 85.88 85.88 85.91 85.86 86.12 86.14 86.26 86.26 86.14 86.14
breast 97.41 97.79 97.44 97.44 97.41 97.41 97.35 97.38 97.18 97.18 97.32 97.32
cleve 84.27 84.27 83.86 83.86 84.07 84.07 83.53 83.53 83.53 83.39 83.05 83.05
crx 85.77 85.92 85.46 85.92 84.85 85.92 84.54 84.38 85.00 85.15 85.00 85.00
diabetes 76.16 76.24 75.37 75.42 75.40 75.40 75.45 75.48 75.35 75.40 74.56 74.56
german 72.86 74.40 72.88 72.98 72.80 72.80 70.56 71.80 71.30 71.94 70.80 71.76
glass2 86.50 86.50 85.63 85.63 85.75 85.75 86.88 86.88 86.00 86.00 86.63 86.63
heart 78.67 78.67 78.44 78.44 77.93 78.15 78.00 78.00 78.00 78.00 77.85 77.85
iris 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33
pima 76.13 76.13 75.74 75.76 76.08 77.03 76.63 76.68 76.78 76.78 76.81 76.81
vote 87.59 87.49 87.59 87.59 87.59 87.59 90.90 90.85 91.22 91.22 91.13 91.13
waveform 82.13 81.87 81.07 82.07 80.67 80.00 82.27 82.20 78.80 81.33 80.00 80.67
Average 84.07 84.36 83.72 83.86 83.64 83.77 83.96 84.05 83.73 83.99 83.72 83.85

Table 2: Comparing MUV(X, <disj,bal>, mle) vs MUV(X, <disj,bal>, js)

2NBs 3NBs 4NBs 2TANs 3TANs 4TANs
(Disj,Bal) MLE JS MLE JS MLE JS MLE JS MLE JS MLE JS
australian 85.51 85.36 85.51 85.51 85.65 85.65 85.51 85.36 85.51 85.51 85.65 85.65
breast 97.15 97.35 97.35 97.35 97.15 97.35 97.15 97.35 97.35 97.35 97.15 97.35
cleve 83.05 83.05 83.05 83.05 83.05 83.05 83.05 83.05 83.05 83.05 83.05 83.05
crx 86.00 86.00 86.00 86.00 86.31 86.31 86.00 86.00 86.00 86.00 86.31 86.31
diabetes 74.64 74.64 74.51 74.77 74.12 74.12 74.64 74.64 74.51 74.77 74.12 74.12
german 71.00 71.20 71.00 72.10 71.00 71.00 71.00 71.20 71.00 72.10 71.00 71.00
glass2 82.55 83.15 84.38 84.38 83.13 83.13 82.55 83.15 84.38 84.38 83.13 83.13
heart 77.78 77.78 78.52 78.52 77.78 77.78 77.78 77.78 78.52 78.52 77.78 77.78
iris 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33 95.33
pima 75.95 75.95 75.34 75.60 75.34 75.69 75.95 75.95 75.34 75.60 75.34 75.69
vote 87.36 87.36 87.36 87.36 87.36 87.36 87.36 87.36 87.36 87.36 87.36 87.36
waveform 80.27 80.73 81.33 81.67 79.33 79.00 80.27 80.73 81.33 81.67 79.33 79.00
Average 83.05 83.16 83.32 83.47 82.93 82.98 83.05 83.16 83.31 83.47 82.93 82.98


How to generate Samples: <2a>  Bootstrap vs Disjoint

Table 3: Comparing Boostrapping and Disjoint samplings for kNBs with JS and Bal

2NBs 3NBs 4NBs
(JS,Bal) Boot Disj Boot Disj Boot Disj
australian 87.68 85.36 85.88 85.51 85.86 85.65
breast 97.79 97.35 97.44 97.35 97.41 97.35
cleve 84.27 83.05 83.86 83.05 84.07 83.05
crx 85.92 86.00 85.92 86.00 85.92 86.31
diabetes 76.24 74.64 75.42 74.77 75.40 74.12
german 74.40 71.20 72.98 72.10 72.80 71.00
glass2 86.50 83.15 85.63 84.38 85.75 83.13
heart 78.67 77.78 78.44 78.52 78.15 77.78
iris 95.33 95.33 95.33 95.33 95.33 95.33
pima 76.13 75.95 75.76 75.60 77.03 75.69
vote 87.49 87.36 87.59 87.36 87.59 87.36
waveform 81.87 80.73 82.07 81.67 80.00 79.00
Average 84.36 83.16 83.86 83.47 83.78 82.98

Table 4: Comparing Boostrapping and Disjoint samplings for kTANs with JS and Bal

2TANs 3TANs 4TANs
(JS,Bal) Boot Disj Boot Disj Boot Disj
australian 86.14 84.64 86.26 85.22 86.14 84.93
breast 97.38 97.21 97.18 97.21 97.32 97.21
cleve 83.53 82.71 83.39 81.69 83.05 82.71
crx 84.38 86.08 85.15 86.31 85.00 85.54
diabetes 75.48 74.64 75.40 73.86 74.56 73.99
german 71.80 69.90 71.94 67.90 71.76 68.90
glass2 86.88 84.38 86.00 84.38 86.63 84.38
heart 78.00 77.78 78.00 77.41 77.85 78.15
iris 95.33 95.33 95.33 95.33 95.33 95.33
pima 76.68 76.34 76.78 76.47 76.81 76.34
vote 90.85 90.11 91.22 89.89 91.13 89.43
waveform 82.20 80.27 81.33 81.33 80.67 77.33
Average 84.05 83.28 84.00 83.08 83.85 82.85


How to generate Samples<2b>:  Balanced vs Skewed

Table 5: Comparing kNBs and kTANs when data samles are either Balanced or Skewed.

kNBs kTANs
Bal vs Ske Disj,Skew Boot,Bal Disj,Skew Boot,Bal
australian 85.80 87.68 85.36 86.14
breast 97.21 97.79 97.35 97.38
cleve 83.05 84.27 81.69 83.53
crx 82.46 85.92 84.62 84.38
diabetes 73.73 76.24 76.86 75.48
german 73.90 74.40 71.90 71.80
glass2 84.38 86.50 85.00 86.88
heart 75.93 78.67 78.52 78.00
iris 95.33 95.33 95.33 95.33
pima 74.12 76.13 78.04 76.68
vote 87.59 87.49 89.66 90.85
waveform 81.33 82.07 80.00 81.33
Average 82.90 84.37 83.69 83.98


How to generate Samples:<2>  Adaboost

Table 6: Comparing MUV(NB,Ada,js) and Adaboost(NB)
Boosting MUV(NB,Ada,js) AdaBoos(NB)
australian 85.23 84.78
breast 97.36 97.09
cleve 83.79 83.35
crx 86.04 79.60
diabetes 76.61 77.17
german 74.01 73.06
glass2 86.42 83.96
heart 78.73 78.60
pima 77.58 76.55
vote 88.76 88.05
waveform 80.82 48.67
Average 83.22 79.16



1