Fig. 6

Result for robustness test on \(d_a\) and \(d_r\). A, B, C, D are results obtained using Pearson correlation as \(\rho\) on PAM. A. The number of times \(d_r\) win over 20 iterations in each dataset. Each row corresponds to one dataset. True difference in the count of win and lose event is not equal to zero (P-value=\(3.41e-14\)). B. P-values in testing the difference between the number of times imply \(d_r\) wins in all 35 datasets. Each point corresponds to one dataset. C. Each box represents one \(\upsilon\) value over 20 iterations per dataset. We compared the box plot for \(d_a\) and \(d_r\) in each dataset. The datasets in C have been reordered to fit the decrease of y value to show the trend more clearly. D. The number of classes “dissolved” in \(d_a\) and \(d_r\) across all 20 iterations. (P-value=0.005). E. Result for Spearman correlation as \(\rho\) in PAM clustering. (P-value=\(1.08e-13\)). F. Result for uncentered Pearson correlation as \(\rho\) in PAM clustering. True difference in the count of win and lose event is not equal to zero (P-value\(=3.39e-13\)).G, H are results for Pearson correlation as \(\rho\) in hierarchical clustering, considering all internal nodes as classes. G. Result for comparing \(d_a\) and \(d_r\) by the number of times classes “dissolved” in 35 datasets over 20 iterations. The number of times \(d_r\) wins, loses, or is equal to \(d_a\). The green horizontal line represents the average number across all the iterations where \(d_r\) wins. The red horizontal line represents the average number across all the iterations where \(d_r\) loses. True difference in the count of win and lose events is not equal to zero (P-value=0.020).H. Result for comparing \(d_a\) and \(d_r\) per dataset. (P-value=0.019)