How to show clusters of events of the same category occuring within spacetime window?

11-08-2012 02:07 PM
New Contributor III

I have a set of points representing crimes, and want to show all clusters of specific crime categories which fall within a space/time window. For example, "show me where there are clusters of vehicle thefts which occurred within the same 1km area in 24 hours". From a crime analysis point of view, the question that I'm hoping to answer is "where have bursts of activity occurred, and which incidents were involved in them, so I can go back to my database and investigate those incidents". I've played around a lot, and looked into the help for Space-Time Cluster Analysis and associated pages, but it's not quite answering my question.

I've used the generate spatial weights matrix tool to specify a 1 km, 1 day neighbour matrix. I've then converted that to a table, which gives me a list of which features are adjacent (in space & time) to which other features. If I run that process once per category, that's my "clustering analysis" complete, really - I don't want to do any statistical stuff from here. Will I have to write my own tool from here to show these clusters on the map (e.g. as MBRs around the points in question)? I guess it's a relatively small step from the neighbour table to such a tool, but I don't want to write something if I don't have to.

I feel like this is either a lot easier than I'm making it out to be, or a lot harder. I have a niggling feeling I should be able to do the same thing with just SQL. If ArcGIS had multidimensional indexing then it could be a 4D buffer/union. Is this a 'hidden feature' that I don't know about? I guess I could hack time into the Z coordinate but that's a bit... well, hacky.

Thanks for your help.

(Cross-posted from the spatial statistics subforum)
0 Kudos
1 Reply
Esri Regular Contributor

If you have your table then you can indeed create bounding hulls for your clusters - see the Minimum Bounding Geometry tool.
Try and keep it simple if you can.

If you feel the urge to get seriously "hacky" there are approaches to multidimensional indexing in space-time but out-of-the-box is a lot easier.

0 Kudos