Finding optimum number of clusters.

<p>The plots represent the comparison of five different methods of finding optimum number of clusters on the dataset generated using moderate effect size of 3.5. First row represents silhouette width over k = 2:8 for each of five different scenarios of true clusters 2, 3, 4, 5 and 6 over 30 runs of simulation. The average value of the silhouette widths over 30 runs are overlaid on the plots as a line. Cophenetic correlation, Dispersion, Residual Sums of Squares and Cluster Prediction Index are shown on second, third, fourth and fifth rows respectively.