Considering time in DBSCAN

0
votes

I want to cluster some GPS points with DBSCAN algorithm and I select eps:20 m and min_samples:4. The algorithm works fine on my data, however I need to also consider time of point in the clustering process. Suppose there are 10 points in a cluster, however 4 of them are between 8 am and 8:30 am and others in duration of 11 am and 11:15 am. What I want is that the algorithm detects 2 clusters here, one with time of 8 points and one with time of 11 points. I mean I need to have another criteria for my DBSCAN algorithm rather than eps and min_samples.

1
pythondbscan
No, you made a 3d point (lat, long, time) and try to cluster based on that - Dani Mesejo
@DaniMesejo I thought about this however I am wondering what would be the unit of distance in a 3d point of (lat,long,time)? since I want the eps to be 20 meters. - mpy
What is the criteria for time? - Dani Mesejo
@DaniMesejo I need the algorithm not only consider the distance between points but also the time of occurrence... the neighbor points should also in a specific time range. - mpy
Yes I understood that, but you said you need 20 meters for the distance between points, what should be the value for that specific time range - Dani Mesejo

1 Answers

0
votes

Use Generalized DBSCAN.

Then define neighbors as being both

  1. within distance maxNeighborDistance km
  2. within maxNeighborTime minutes