Training Classifiers under Covariate Shift by Constructing the Maximum Consistent Distribution Subset

Table 1

Comparing ICD1 with SVM and TrAdaBoost.

Data set	SVM	IDC1	TrAdaBoost
Australian	0.756 ± 0.082	0.889 ± 0.052	0.836 ± 0.055
Balance	0.721 ± 0.072	0.846 ± 0.060	0.803 ± 0.062
Breast	0.698 ± 0.088	0.779 ± 0.086	0.722 ± 0.082
Cleveland	0.623 ± 0.072	0.697 ± 0.067	0.660 ± 0.061
Credit	0.792 ± 0.020	0.902 ± 0.025	0.826 ± 0.031
Diabetes	0.702 ± 0.056	0.816 ± 0.046	0.733 ± 0.039
Heart	0.729 ± 0.092	0.832 ± 0.070	0.830 ± 0.072
Ionosphere	0.837 ± 0.053	0.926 ± 0.052	0.903 ± 0.059
Iris	0.805 ± 0.061	0.976 ± 0.044	0.915 ± 0.049
Liver	0.672 ± 0.062	0.729 ± 0.038	0.732 ± 0.042
Page	0.892 ± 0.015	0.982 ± 0.021	0.901 ± 0.022
Sonar	0.705 ± 0.105	0.820 ± 0.102	0.767 ± 0.099
Thyroid	0.825 ± 0.081	0.961 ± 0.076	0.882 ± 0.075
Vehicle	0.708 ± 0.021	0.798 ± 0.025	0.790 ± 0.026
Voting	0.825 ± 0.046	0.963 ± 0.022	0.896 ± 0.028
Waveform21	0.721 ± 0.026	0.823 ± 0.017	0.821 ± 0.010
Waveform40	0.772 ± 0.025	0.838 ± 0.019	0.786 ± 0.029
Wine	0.793 ± 0.098	0.915 ± 0.092	0.898 ± 0.109
wdbc	0.803 ± 0.036	0.966 ± 0.023	0.909 ± 0.018
wpbc	0.682 ± 0.113	0.775 ± 0.099	0.726 ± 0.094