0
$\begingroup$

I'm wondering if multiplying the feature values by 100 would have any impact on the clustering results of K-means, Hierarchical clustering, and DBSCAN clustering. Suppose I'm using Euclidean distance. For Euclidean distance I assume the results won't change? How about using Manhattan distance?

$\endgroup$
0
$\begingroup$

For DBSCAN you clearly have to also adjust the epsilon parameter.

But apart from that it is trivial to prove that for any Minkowski metric (including Euclidean and Manhattan) the distances increase by a factor of alpha if you scale all attributes with alpha.

For k-means the situation is slightly different. It does not minimize Euclidean distance (not Manhattan) but it minimizes Variance. Variance is squared, so all variances grow by alpha². Nevertheless, the solution found by k-means will not change because all terms just change by this constant (again, this is trivial to see and prove).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.