topic Re: Exclude points before finding mean center in Spatial Data Science Questions

Exclude points before finding mean center

sotokan80_ — Tue, 06 Mar 2018 08:06:45 GMT

Is there a way to exclude points that follow a certain pattern before finding mean center. (i.e in the image below i want to exclude certain wrong gps points, yellow circles, that will affect my mean center).

Re: Exclude points before finding mean center

DanPatterson_Retired — Tue, 06 Mar 2018 09:33:57 GMT

The easiest solution is to get the selector tool, and draw a rectangle around the points you want to include, then run the Mean center tool. Only selected features are used during any arctoolbox tool executation

Re: Exclude points before finding mean center

sotokan80_ — Tue, 06 Mar 2018 10:03:24 GMT

thank you very much for your reply, but i forgot to mention that this is just an example. I have to repeat the same procedure for thousands of places, so i m trying to figure out a way to exclude outliers.

Re: Exclude points before finding mean center

DanPatterson_Retired — Tue, 06 Mar 2018 10:17:38 GMT

more complicated if you don't want to provide the input.

What has worked in the past for me was

calculate the mean center for the point cluster
calculate the standard distance
remove any points beyond some threshold (perhaps 3 standard distances)
recalculate and repeat if necessary

Alternatives

use median center ! probably the better measure of centrality anyway
do a k-means clustering as the basis for your clusters
sort the data, perform a 'trim mean' by removing the top and bottom X% (ie remove points from the tails of the ordered pattern

There are lots of alternatives, but a better description as to why there are thousands of places and do they exhibit the same phenomenon that causes outliers in the gps data.

If it is the gps data collected within a particular time frame from a stationary location, then calculate the temporal difference in the time sequence and remove the points that exhibit anomalous dx, dy properties in the time sequence.

A quick measure in the first instance.... Median center, first, then provide your rules of refinement as your 'corner-cases' arise