Please take a look at my accepted answer to another question about PCA. It is effectively a tutorial on using PCA in MATLAB.
You don't explain much here, but I'm guessing the last column of your data is the number you are trying to predict, using the other columns as predictors?
Doing a PCA on the first 39 columns does look like a good idea. The first two principal components capture over 99% of the variation.
Then you could do a logistic regression (perhaps using fitglm) to predict the last column from the first two principal components.