Clustering algorithm - how can there be new clusters created if the clustering algorithm is k-means?

[Bernard Esterhu]

We are making use of the Clustering feature in Predictive Intelligence. We are using the default clustering algorithm, which according to the documentation is k-means.

Currently we have our Solution Definition as:

Update Frequency: Every Hour

Training Frequency: Run Once

My understanding is that the initial training chose the number of clusters (k), and the update step adds new incidents to existing clusters. However, if I look at the Solution->Cluster Updates, I see some new clusters created. How does this work? Is my understanding of "Training" and "Update" incorrect?

[Lener Pacania1]

You're correct Bernard.  When you initially train the clustering model, we internally calculate the optimal K and then build the clusters.  Your settings "update frequency = every hour and training frequency = run once" this should not create new clusters (k) and should only add the new records to the existing clusters.  It's been a few months since you posted this is it still happening?  In the future, for faster responses on Predictive Intelligence questions please post to the AI community.

[Bernard Esterhu]

@Lener Pacania1 thanks a lot for the update.

We discussed this with the ServiceNow engineers in the ML team, and the info they gave us was that with the default k-nn model, new clusters will be formed in time for incidents that do not fit in the clusters created during initial training. This seems to be different from what you explained above?