hansieh

ArcGIS Pattern and Cluster Analysis Tools

Discussion created by hansieh on Jun 6, 2012
Latest reply on Jul 16, 2012 by dlee-esristaff
I want to categorize all the landmines in a minefield into one cluster or another.  I do not want to use anything related to statistical significance for doing so, e.g. z-scores, p-values.  Rather, it is more of a classification problem than a hotspot analysis.  Let me try to explain:

Using a 1-dimensional analogue, consider:

      x  x    xx   x      xxxxx   xxx

If we need at least two x's to be in any given cluster,  the above might be considered as 4 clusters (sizes 2,3,5,3 going left to right) or three clusters (5,5,3) or 4 clusters with an outlier (2,2,outlier, 5,3).  An algorithm might try putting these together in various ways, each time calculating the ratio of between-cluster variance to within-cluster distance variance, and choose the categorization that
maximizes this ratio.
Does anybody knows if ArcGIS has a procedure that could figure out for a given minefield what are the best categorizations that result in 2,3, or 4 clusters, then choose the number (2,3,or 4) that is 'best' in some sense, then labels each mine as to being in which cluster?

Outcomes