I'm trying to classify/cluster subjects according to 4 features in two classes: healthy and sick.

Two things to know: I know the labels/classes of each subject + I only have 40 subjects (in total: training + testing set!)

What should I choose in this case, clustering or classification?

  • $\begingroup$ Classification, it's clearly a supervised problem. If sample size is an issue, consider doing statistical tests on the full data set (e.g. with logistic regression) rather than using a training and test set. $\endgroup$ – Denziloe Aug 21 '18 at 20:24
  • $\begingroup$ I'm not really familiar with logistic regression, can you elaborate a little more please? $\endgroup$ – learneRS Aug 21 '18 at 20:58
  • 1
    $\begingroup$ I can't go into much detail, that's for you to research. But in a nutshell, logistic regression will try to model the probability of health/sickness via a linear sum of your 4 features. Because it's a statistical model, instead of gauging accuracy with a test set, you could instead do hypothesis tests or get confidence intervals for the coefficients in the linear sum using statistical software. In Python for instance you can use this: statsmodels.org/dev/generated/… $\endgroup$ – Denziloe Aug 21 '18 at 21:02

If you have labels, it's classification.

Clustering would be unsupervised. You can't ask it to classify into healthy and sick, it may find a very different partitioning of your data of that provides more structure.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.