3.8 LDAT performance on the dataset in Figure 2.2(a) with different
neighborhood size k. k is set as the percentage of n/c (n is #instances
and c is #clusters). AHK+LDAT has better and more stable
performance than RWC+LDAT as k changes. . . . . . . . . . . . . 53
3.9 Stability with different adaptive scaling parameter q. . . . . . . . . 59
3.10 Stability under different neighborhood size k. . . . . . . . . . . . . 60
3.11 Stability under different reduction factor α. . . . . . . . . . . . . . 61
3.12 Algorithm performance on different noise levels. . . . . . . . . . . 62
3.13 Scalability Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . 62