Important predictors in PCA analysis and Pareto

friends, I have a data set of (1400*4); 4 parameters (mass, area, distance, color) are measured for a sample size of 1400. In order to find the important predictors, I used PCA command after the data were normalized. [pcs,scrs,~,~,pexp] = pca(statsNorm); Then I plotted the Pareto chart ( pareto(pexp)) and the labels beneath each par are just (1, 2, 3 and 4). How can I figure out the correspondence between these numbers and the predictors? In other words, what is 1! mass, area, distance or color?

Answer 1

[the cyclist]

The first bar corresponds to the first "principal component". It will not be one of your original variables, but rather the linear combination of your four variables that captures the largest amount of total variance.

This may sound unkind, but your question shows that you don't under what PCA is intended to do. I don't mean the MATLAB pca command, but rather the technique itself.

Rather than explain the whole technique here, and what it is useful for, I would encourage you to read a simple article about it -- maybe even just Wikipedia.

What are you trying to predict? PCA, by itself, is not really a predictive modeling technique.

[Tala Hed]

you may be right since I am very new to machine learning but I am trying to cluster my data, no prediction yet. I get a plot after I employ the following commands: [pcs,scrs,~,~,pexp] = pca(statsNorm); grp = kmeans(statsNorm,2,'Replicates',5); gscatter(scrs(:,1),scrs(:,2),grp,'br','xo')

What I get is a plot of two clusters and I have no idea what are the axis in the plot.

[the cyclist]

I'm afraid that it is just not very practical to give you an entire tutorial on the meaning the principal components, and their relationship to the original data. If you are going to use this technique, you really need to learn it from some other source than me.