How to perform unsupervised Spatial Clustering of Polylines

Question

I'm trying to group polylines (running routes) based on their similarity - any guidance on the best tool and starting point?

Specifically, I have 600 or so running routes, each with a spatial object and total distance.

I'd like to group routes based on the most common tracks that I run.

Yoshiro_Fujimori · Answer

Thank you for the information.

I am not an expert of Clustering at all,

but I guess the issue is how to apply K-Centroid Cluster Analysis tool in Predictive Grouping to your case.

If you are not familiar to this tool, check the interactive lesson;

https://community.alteryx.com/t5/Interactive-Lessons/Clustering-in-Designer/ta-p/679556

The problem is that the tool accepts only numeric fields as the input data field (not spatial objects).

So one idea is that you input the following fields to the tool and see if it works.

ST_CentroidX([startPoint])

ST_CentroidY([startPoint])

ST_CentroidX([endPoint])

ST_CentroidY([endPoint])

ST_Length([route],"Mi")

You may want to add the data features, say

ST_NumPoints([route])

ST_CentroidX([route])

ST_CentroidY([route])

ST_MaxX([route])

ST_MaxY([route])

ST_MinX([route])

ST_MinY([route])

If you want more, you may want to add

1st quarter point

Halfway point

3rd quarter point etc.

(But it may require a little more efforts of working on Spatial tools. I would not deep dive it for now.)

I hope this helps. Good luck.

chrismoshier · Answer

Thanks for the quick reply. Here's an example of 6 routes in Chicago. They're all a little bit similar, but only 4 and 5 are "similar" based on the actual distance, the starting/ending point, and the actual route.

Yoshiro_Fujimori · Answer

Hi @chrismoshier

Could you describe what you think as "similar"?