Skip to main content
. 2016 May 26;7:774. doi: 10.3389/fmicb.2016.00774

Figure 4.

Figure 4

A comparison of the 10 clustering algorithms with the Calinski-Harabasz validity index (CH index) for solutions containing between 2 and 10 clusters, where higher index values indicated better intra-cluster cohesion and inter-cluster separation. Of these clustering methods, the partitional methods fuzzy c-means (dashed green) and k-means (solid orange) with five clusters were found to be the optimal algorithms for the isotope and elemental ratio NanoSIMS data.