First time here? Checkout the FAQ!
x
+1 vote
551 views
asked in Machine Learning by (660 points)  
how to choose the number of clusters based on intertia and based on silhouette using K-means
  

1 Answer

0 votes
answered by (115k points)  
Having these different approaches to determine a convenient value of k (number of clusters) should help to narrow down the options for $k$, however, we cannot make sure the value we select finally is the best or not. For example, if silhouette measure is closer to 1 when $k=5$ and for Inertia we have the "elbow" or "knee" around $k=4$ to $k= 6$, probably one of the values of 4, 5, or 6 might be a good value for the number of clusters. These two methods together help us have some estimations for the number of clusters but they do not guarantee it.
commented by (660 points)  
That makes sense. But I am wondering if you can guide me on what is silhouette method and what is elbow method
commented by (115k points)  
Please check out the following links:
http://scikit-learn.org/stable/auto_examples/cluster/plot_kmeans_silhouette_analysis.html
and for inertia and see how you can choose it based on "knee" or "elbow", please check the following link:
https://stackoverflow.com/a/41525972
...