Figure 2

Learning curves of aggregating PCDA and PCDA. With the increasing of sample size for 12 to 100, the classification performance of aggregating PCDA and PCDA is increased significantly, and the variation of classification models also tends to be reduced. The classification performance is represented by 1 minus misclassification rate, and the variation of classification performances is represented by an error bar. The upper error ranges for each point in error bar is obtained with adding standard deviation of mean of classification performance and lower error ranges is obtained with subtracting standard deviation of mean of that. The figure shows that aggregating PCDA usually gives a better classification performance than PCDA. The classification of validation sets and test sets are quite similar since two sets follow the same distribution.