Comparing Clustering Over Time

Question

I've recently conducted a k-prototypes R routine on some mixed data.

In particular, the data is health data concerning a certain public health intervention, with categorical variables for health scores and numerical demographic data such as age

The utility scores were measured at different points in time, with one sample at week 1, and one sample at week 10.

At the moment, I have only conducted clustering analysis on one of these samples. However, I wondered if there is a recognised routine for making inference on the clusters between week 1, and week 10.

All feedback would be appreciated. I recognise that it is not quite longitudinal clustering, but more comparing two different clustering states.

Thank you

Answer 1

There are many options. One option could be to find clusters at the first time point and define a change "metric" at the second time point. A possible change metric is if the data points belong to the same cluster at the second time point.

You mention a variation of k-means. Since k-means is an iterative algorithm, this cluster re-assignment is what happens between each training step. The difference with this method is you are allowing the data points to move between training steps and measuring how many cluster assignments change.