I had a question in regards to PCA with times series data, and specifically how to possibly interpret it. Normally, PCA is used by other software that I use in relation to de-noising a data set by removing components of least eigenvalues/variance (not sure as to the precise criteria used here), and the de-noised data is passed along for other preprocessing steps.

However, I decided to manually use PCA on my data and actually get a look at the results of doing so. Essentially, I have 5 variables (5 brain regions), each with 180 time points, with specific intensity values (unit is same across the 5 variables).

Now, my experience with PCA is limited, but I was wondering if there are any potential interpretations that could be done with such a time series data set?

From what I understand, the 4 variables (NOT the Cerebellum) are highly related to the PCA axis 1, and these 4 (RightPM, RightM, LeftPM, leftM) can be regarded as positively correlated with each other in relation to the intensity values? Confusion sets with the time points however, such as saying whether this strong positive relationship occurs at the time points plotted near the bunching of the 4 variables.

I'm assuming there are issues with PCA and time series (related to autocorrelation perhaps), but I was curious if such an interpretation could be made. Any opinions on this would be great.

Thank you.

PCA Biplot of 5 variables with each having 180 timepoints

  • $\begingroup$ Your interpretation of the variables ("4 variables ... are highly related to the PCA axis 1") is more or less correct, but your second interpretation is completely wrong: "this strong positive relationship occurs at the time points plotted near the bunching of the 4 variables" -- not true (and it is not even entirely clear what this could mean). $\endgroup$ – amoeba says Reinstate Monica Dec 29 '14 at 13:51

Your Answer

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

Browse other questions tagged or ask your own question.