Hello everyone,
I have basic longitudinal data of anthropometric features (weight, waist size, height…) over a given period of time.
I’m trying to isolate patterns/clusters/trends of evolution in this longitudinal data.
What approaches are efficient with these small time series of numeric features ?
Cheers and happy coding,
Try to check out the Machine Learning Specialization, they explain the clustering algorithm there!
Not my area of competence, let alone expertise, but I think time series require some adaptations of the common clustering techniques that rely on Euclidean distance and ignore the time dimension. I don’t recall that MLS covers that nuance.
Seems to depend on whether the objective is to compare and cluster entire time series or merely the values they contain? Might have missed it in a quick read, but I don’t see time series mentioned in the documentation for algorithms from common toolkits like scilearn, so be cautious that the optimization approach they use is appropriate.
Articles like this might provide ideas…
A review and comparison of time series similarity measures
Maybe look into Dynamic Time Warping. See for example
Summarizing a set of time series by averaging: From Steiner sequence to compact multiple alignment
François Petitjean, Pierre Gançarski
Let us know what you find?
you can use k clustering algorithm but one needs to know if your numeric features follow what kind of correlative analysis, linear or logistic.
then you can use scikit data analysis clustering too.