Hi,
I am a botanist and in my experience, LDA always finds better differentiation than PCoA among a-priori defined groups (for me = taxa). I always use PCoA for molecular datasets but for morphology LDA suits better my expectations about species differentiation. My question is, by using LDA, am I inflating my own prejudices about the groups or is it legitimate to do so? What are the limits of LDA? Thank you.
Remember the goal is to understand the data not to agree with your expectations. Best regards, David Booth
First of all, PCA is an unsupervised method wich can by applied to a set of variables X while the LDA is a supervised one with a dependent variable y to predict. Now , if you face the problem of the curse of dimensiality, you can start by extracting the first principal components from X and then perform a LDA of y on these components
Hi, thank you for your interest.
My question, more simply formulated is: in my experience, LDA offers better discrimination among groups than PCA. Is this normally so or just an artifact? Is ist legitimate to use always LDA for group discrimination or there are limits?
Thanks for relevant answers :-)
- Badges
- Science topic