Spatial clustering with profiling variables/weights



I have a list of demand points in an area, with latitude, longitude, revenue potential and other variables.

Tried k-means clustering to group these points using Lat, Long and Revenue. The idea is to account for both geographical contiguity and profile. Is there any other way to approach this with spatial statistics? ArcGIS has an extension called Spatial Analyst, SAS has SOM and Alteryx Gallery also has an alternate macro to generate weighted centroids. Has anyone tried these? Or any suggestions on trying this in R?

Thanks in advance !


Hi @kapooraparna

Why do you want to use kmeans? if you want to group by region and profile I shall use hierarchical clustering with a tailored distance function. I can imagine you do a distance type euclidian for the long lat give a weigth for the final calculation and then add (or other function) the Revenue (knowing that the distribution should be skewed you could do a transformation before). Then you build the distance matrix and then do a hierarchical clustering that you prune.

Any body with better suggestion(s)?

Have a good day