How should we place the clusters in a K-means clustering implementation?

machine_learning
k-meansclustering
data_science

#1

Hello,

After deciding the number of clusters we want, how should we place the clusters so that the algorithm converges closest to the global optimum solution?
Should we just randomly initialize the clusters or should we make sure that the clusters are equally distant from each other? What is a good way?

Thanks!


#2

Hi Aditya,

Random initialization is your safest bet in this case. What you can probably do is repeat the clusteting 10 time to reach to an optimum solution after looking at intra cluster and inter cluster distance.

Regards,
Aayush