I have created a PCA plot for a dataset which has 6 groups. Post PCA I find very high values of PC1, but the groups are not clearly separated in the image.

What could be the reason? Is there some other metric than PC1 values to check for group separation using PCA

enter image description here

  • 3
    $\begingroup$ Principle components account for variation contained in the data. PC1(98.88%) implies that 98.88% of all the variation in the data can be explained by PC1. In other words, there isn't all that much variation here if just one principle component can capture this much of it. $\endgroup$ May 5 '20 at 17:20
  • $\begingroup$ But coming to your point regarding separation, wherever two groups are close, this means that you can expect them to be not all that different. But there is indeed clear separation of groups a and f. $\endgroup$ May 5 '20 at 17:26
  • $\begingroup$ @AnuragN.Sharma even if the group a and f is far but the group b, c and d are close. Won't this affect the overall score? or this has nothing to do with the PC score $\endgroup$ May 5 '20 at 17:38
  • 1
    $\begingroup$ The distance between the groups implies the variation that exists between them along that particular PC which is PC1 in this case. So you might summarise your figure as: PC1 accounts for 98.88% of the variation in the data wherein groups a and f show the biggest difference amongst all group-wise combinations. $\endgroup$ May 5 '20 at 17:42
  • $\begingroup$ This doesn't appear to be a specific programming question that's appropriate for Stack Overflow. If you seek recommendations for statistical methods for clustering observations, then you should ask such questions over at Cross Validated instead. You are more likely to get better answers there. $\endgroup$
    – MrFlick
    May 5 '20 at 19:25

What could be the reason? This means that the first two PCs don't separate the groups.

PCA is an algorithm. This means that PCA has no information about the groups you want to separate. By contrast, is explicitly designed to separate groups according to their features. You would probably have more success using a supervised method.


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.