Mckinsey Interview Questions | Clustering Algorithm
Question
How will you define the number of clusters in a clustering algorithm?
(Hint- Try to be specific)
in progress
0
Machine Learning
55 years
2 Answers
921 views
Great Grand Master 0
Answers ( 2 )
The elbow plot gives us the WSS (With-in-Sum-of-Squares) value against
the number of clusters. WSS is the total distance of data points from their
respective cluster centroids. Although, there is no ideal answer for this question, it is
suggested that you should choose the no of clusters after which point you stop
seeing a reduction in the WSS.
In the figure above, after k=6, we don’t see much reduction in the RSS.
So, we can choose 6 as the optimal no of clusters.
Three main techniques for finding the number of clusters in any cluster algorithm
1. Elbow method
2. Average Silhouette method
3. Gap statistic method