www-ai.cs.tu-dortmund.de/LEHRE/FACHPROJEKT/SS12/paper/outlier/marascu2009.pdf
ignore clusters with equal size. For instance, with s = {1, 1, 2, 2, 3, 5, 10} a series of sizes and k = 3, the top-k filter will select the 3 first clusters having sizes: 1, 1 and 2, but will ignore the 4th [...] For instance, with s = {1, 3, 10, 11, 15, 20, 55, 100} a series of sizes and x = 0.1 we have d = 100 − 1 = 99, y = 1 + (0.1 × 99) = 10 and the set of outliers will be o = {1, 3, 10}. In our experiments, [...] outliers ranging from 24 (1 %) to 70 (9 %).
Figures 3 and 4 give a comparison of WOD (applied to the same data) with top-k and percentage filtering. In figure 3, we compare WOD with a top-10 and a top-80 filter …