Suppose we have an unlabeled dataset containing {x1, x2, x3,…, xn}. We run KNN with 50 random initialisations and obtain 50 clusterings of the data. How do we choose which cluster to use?
We would need the yi values to be able to predict this.
Compute the inertia corresponding to the xi and use the one which maximises it.
Compute the inertia corresponding to the xi and use the one which minimises it.
Choose one of the later clusters since by that time it must have converged to a good solution
Any of the above