S-RASTER: contraction clustering for evolving data streams

Journal of Big Data

Table 2 Comparison of SW-RASTER with various standard clustering algorithms

	SW-RASTER	Windowed k-means	DStream	DBStream
Macro clusters	4	4	3	2
Micro clusters	103	100	108	118
purity	0.93	0.94	0.96	0.97
SSQ	77.80	114.26	50.70	44.72
cRand	0.04	0.06	0.06	0.06
silhouette	0.18	0.18	0.21	0.27
Manhattan	00.11	0.12	0.13	0.13

The best values in this comparison are listed in italics. The number of micro clusters is listed for the sake of completion but we abstain from making a judgment as the resulting macro clusters are more relevant. Our algorithm does well in this comparison, as evinced by the cRand, silhouette coefficient and Manhattan distance values