The Data Mining Forum (v2)

Hello
I need to answer this question for my assignment:

What are the main limitations of K-means?

I think the key issue is how to set the parameter k. Is there some other limitations?

thanks in advance

also kmeans tend to find clusters having a spherical shape, dont work well with clusters of different shapes and densities

About the parameter k, first it Is hard to set and second there Is the problem of cluster initialization. I mean that depending on how the cluster centers are initialized can give very different results.

gmc wrote: ↑Wed Apr 06, 2022 11:31 pm About the parameter k, first it Is hard to set and second there Is the problem of cluster initialization. I mean that depending on how the cluster centers are initialized can give very different results.

thanks All for your answers. That means a lot to me

The main limitations of K-Means are that it needs the number of clusters as an input, it is a biased algorithm, and also it fails to detect outliers. Outliers can be classified as any data that are not near the cluster or do not belong to the cluster. These outliers have a tendency to break down the dataset, thereby rendering the accuracy and performance of the K-Means algorithm weak.

Some important limitations of K-means are that it is sensitive to outliers, it requires the specification of the number of clusters, and it tend to produce clusters that have globular shapes.

Also K-means is not suitable for clustering data with non-numeric features or categorical data, as it relies on the mean of the data points in order to determine the clusters.

Finally, K-means can fail to converge on the optimal clustering solution if the initial cluster centers are chosen poorly.