For your problem, if your dataset has a large number of features, then you can use Principal Component Analysis (PCA). It is a dimensionality reduction technique, which means it chooses highly co-related features based on the variance between them. Once you got k numbers of feature, you can define your k-means cluster value based on that.
Another approach to your problem is cross-validation. You choose a subset of data and k-means will make k number of clusters. Then you can compare clusters of a subset with the rest of the data.
K-means halts creating and optimizing clusters when either:
from sklearn.cluster import KMeans
Kmean = KMeans(n_clusters=2)
Hope this answer helps.