Principal Component Analysis

shortys94

Hi

I am trying to analyze 4 groups of variables, 3 groups represent the geometrical attributes of a patient and are independent variables. 1 group are dependent variables which consist of the dose to a patient.

All of the data are observations made by myself. I want to know how do I use PCA to illustrate how the variation in the independent variables changes the dependent variable.

I have done this using R, I used matrix to input the data in 4 columns, but did not separate the dependent and independent variables from each other, I just created the matrix with 4 columns of 45 rows. I managed to calculate the PCA but only have a scatter plot which has the PCA 1 as x axis and PCA2 and y axis, but because I did not signify a dependant variable the data just seems like a scatter plot with no meaning, only the number of each patient noted on each data point.

Does anyone have any advice how to go about PCA for a situation like this?

Thanks

shortys94

---

PC1 PC2 PC3 PC4
Dose of A 0.60304092 - 0.187254393 0.41822640 0.6529656
Target Volume 0.48414567 -0.007968846 -0.86840797 0.1068039
Volume of A 0.04588944 0.975655799 0.04248323 0.2102025
Overlap Volume of A 0.63232806 0.113877358 0.26296285 -0.7197525

Above are my principal component loadings, again the dose is dependent did not specify this in R, just put my matrix in

shortys94

Cant format properly right now, i dont know if its clear

defrule

When you get your PCs out, there will also be a cumulative proportion of variation explained in the output. Say for example with PC1 and PC2 we can explain over 80% of the variation then we would choose to use those two PCs. Therefore we reduced the problem from 4 variates that maybe be correlated into 2 PCs that are orthogonal.

Every observation you have will have the PC1 and PC2 score and that scatter plot will help you identify the groups and outliers.

You should be able to interpret the PCs based on their loadings on the original variates.