Algorithm Number of clusters Neighbourhood size (Chameleon only) Number of sub-partitions (Chameleon only) link purpose Examples results in figure: result
Chameleon 15 15 - 1000 30, 60, 90, 150 single Assess performance under different combinations of neighbourhood size and number of sub-partitions 3 chaining increased with increasing number of sub-partitions over 30, mis-classification rate increased with neighbourhood size, no clear patterns in within-cluster homogeneity with either neighbourhood size or number of sub-partitions
Chameleon 15 15 - 1000 agglomerative phase omitted NA Assess performance under different neighbourhood size with no agglomerative phase 4 mis-classification rate decreased strongly,within-cluster homogeneity decreased weakly with increasing neighbourhood size
Chameleon 15 30 30, 60, 120, 180, 240, 300, 400, 500 complete Assess effect of increasing number of sub-partitions on a 15-cluster solution with neighbourhood size of 30 samples 4 mis-classification rate decreased weakly, within-cluster homogeneity decreased strongly with increasing neighbourhood size
Chameleon 15, 30, 60, 90, 120, 150 200, 250 30 agglomerative phase omitted NA Assess performance on cluster solutions of different thematic scale, neighbour size fixed, agglomerative phase omitted 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively even in size
Chameleon 15, 30, 60, 90, 120, 150 200, 250 30 15, 30, 60, 120, 180, 240, 300, 400, 500 complete Assess performance on cluster solutions of different thematic scale, neighbour size fixed, number of sub-partitions proportional to number of final clusters 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively even in size
Chameleon 15, 30, 60, 90, 120, 150 200, 250 1000 30, 60, 120, 180, 240, 300, 400, 500 complete Assess performance on cluster solutions of different thematic scale, neighbour size fixed, number of sub-partitions proportional to number of final clusters 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively even in size
k-means 15, 30, 60, 90, 120, 150 200, 250 NA NA NA Assess performance of k-means algorithm over cluster solutions of different thematic scale 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively even in size
flexible unweighted pair-group averaging with arithmetic mean (Belbin et al. 1992) 15, 30, 60, 90, 120, 150 200, 250 NA NA complete Assess performance of agglomerative algorithm over cluster solutions of different thematic scale 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively uneven in size
polythetic-division (MacNaughton-Smith et al., 1965; Belbin et al., 1984) 15, 30, 60, 90, 120, 150 200, 250 NA NA complete Assess performance of divisive algorithm over cluster solutions of different thematic scale 5,6,7 mis-classification rate and within-cluster homogeneity increased with increasing thematic detail, cluster solutions relatively uneven in size