k-Distance Plot for DBSCAN parameter determination (epsilon)

Hi to all,
I am using the DBSCAN algorithm and I want to determine the values of MinPts and Epsilon.
For the MinPts there is a theoretical way to find it, meanwhile, for determining the Epsilon there is another way by using the NearestNeighbors algorithm, to calculate the average distance between each point and its n_neighbors. Then you sort distance values by ascending value and plot to find the elbow in the plot. It should be something like this

2021-04-30 18_10_34-DBSCAN Parameter Estimation Using Python _ by Tara Mullin _ Medium

How can I implement this in Knime? Did somebody use this technique to determine epsilon while using DBSCAN? Which nodes should I use?

Thank you in advance,

There is pretty close to DBSCAN OPTICS algorithm. It has a graphical output, so you can play with Epsilon.

@izaychik63 thank you for your response. Do you mean the node Optics Cluster Compute? I don’t see any graphical output here or am I checking the wrong node?


What’s New in KNIME Analytics Platform 3.5, KNIME Server and KNIME Big Data Extensions | KNIME


@izaychik63 Thank you :grinning:

@izaychik63 I tried this and it works, although I can not order the distances of the graph. Have you used it for determining the epsilon? Any idea how you chose the best epsilon?

As I worked mostly on outliers, I try until get the number giving me expected number of them.

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.