I'm working on an assignment which involves two sets of data, each a matrix whose columns are signals. The signals represent the observed knee motion angles of individuals walking - i.e 0 represents a "straight" leg, and 90 a leg bent at a 90 degree angle. Both sets of data are randomized in a way known only to the instructor. One set contains observations for individuals with arthritis, and one set contains "healthy" individuals - the control group. My task is to determine which is which using principal component analysis (in MATLAB).

I've performed the PCA but I'm wondering if my interpretation of the results is correct. The cumulative variance explained by the "first" five principal components for the first dataset ($X$) is:

$$[0.6582, 0.8151, 0.8927, 0.9480, 0.9802].$$

So the first principal component explains 0.6582 of variance, the first two 0.8151, and so on. For the second dataset ($Y$):

$$[0.7033, 0.8532, 0.9545, 0.9815, 0.9925].$$

My hypothesis is that the control group without arthritis will have greater variance between observations than the group with arthritis, who will have poor range of motion. As such, fewer principal components should capture more variance (from the mean signal), for the "arthritis" dataset, relative to the control dataset. That would make $X$ the control dataset.

I'm not sure if my reasoning is correct or if I've approached this wrong.

  • 1
    $\begingroup$ Your reasoning is not ungrounded but without actually seeing a plot of the components it will be hard to draw better conclusions. Please also add the [self-study] tag as it is clear an assignment. $\endgroup$
    – usεr11852
    Commented Mar 4, 2016 at 23:41
  • $\begingroup$ Your hypothesis has an appeal. Have you been provided enough detail to make hypotheses about how different variables should change? (if not some searching can help) then look at the loadings and see if the pattern you observe matches what you expect. Remember loadings have arbitrary direction (+ or -) , but the scores for that loading will be determined by loading direction in the model. $\endgroup$
    – ReneBt
    Commented Mar 1, 2020 at 7:18

1 Answer 1


Since your data set is clearly random(as only the instrcutor knows it), then the results must be similar. However, you do not know whether data


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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