I want to perform a regression using neural networks. My input has 5 parameters [a, b, c, d, e] and the output is 4 variables [x, y, z, w]. Total number of observations I have is 1000.
I wanted to know if there is a way to know which of the input parameters causes the most variation in my output variables. I thought of doing a PCA analysis where I create a matrix of one input parameter with the output variables (e.g. [a, x, y, z, w], [b, x, y, z, w] and so on) to see if the input parameter can express any variability in the output variables. I didn't quite get any such information.
So my question is:
- Is my approach correct?
- If not, is there any way I can do such a study?