The convergence of PMT-UC with respect to different initializations.

The convergence of PMT-UC is explored by considering the selected parameters and , and the experiment results RI, aRI, SHD, FN and TN, with respect to different initializations using K-means. SHD1 and SHD2 are the results for the first and second clusters respectively, FN is the number of informative variables incorrectly selected to be noninformative and TN is the number of noninformative variables correctly selected. In the true case, , .§c(d): c and d are the average and standard deviation of corresponding results in 10 experiments with a fixed dataset, respectively.