Cluster robustness according to the assumed number of clusters in the dataset (using the PAM algorithm with Manhattan distance).

<p>The cluster robustness evaluates the stability of groups while iterating the same clustering method with the same parameters, except for the assumed numbers of clusters in the dataset (from 2 to 6). The black line represents the median value, the bottom and top of the box represent the 1^st and 3^rd quartiles, and the ends of the whiskers represent minimum and maximum values. Highest median robustness with lowest dispersion was achieved considering 4- and 5-way classification. The examination of the relations between 4- and 5-way classifications revealed that while three groups were constantly retrieved in both cases, the fourth group in the 4-way classification split into two when the 5-way classification was applied (see <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0119161#pone.0119161.t002" target="_blank">Table 2</a>).