Figure 5: Trends in mis-classification rate and within-cluster homogeneity with increasing neighbourhood size or increasing number of sub-partitions in the agglomeration phase. The effect of increasing sub-partitions using the single-linkage function is not shown due to chaining as described above). Trendlines are least-squares linear regressions. Data describing the respective 15-cluster solutions derived by k-means, agglomerative and divisive algorithms are plotted for comparison (see figure 6).