808137_en-US

liliconcom

Creating predictive models for correlated and high-dimensional data

Learn to fit models using variable selection techniques, including shrinkage techniques specifically geared toward modeling correlated and high-dimensional data. Learn how to use the to explore a variety of distributions for responses that are continuous, binomial, count, or zero-inflationary, and how to compare models built using other techniques.

This video is currently being processed. Please try again in a few minutes.

(view in My Videos)

Advanced Statistical ModelingRe: Creating predictive models for correlated and high-dimensional data

There was still a question regarding the integrability of quadratic effects and interactions into the PCA regression:

Theoretically, the relevant columns would have to be created with formula columns before the principal component analysis (for example, column 1*column 1 for a quadratic effect) and then also taken into account in the principal component analysis. Depending on the number of columns, a lot would have to be added here. In addition, multiplying the columns would distort the scaling.

If such use cases exist, PLS with JMP Pro is the more direct solution.

Re: Creating predictive models for correlated and high-dimensional data

Hello Jonas-

Thanks for the answer. I was able to answer the question myself. It is possible, for example, with 2 factors, to model the resulting Prin1 and Prin2 using "Fit Y by X/Selection of RSM" using principal component analysis, the resulting terms Pin1, Prin2, Prin1xPrin2, Prin1xPrin1, Prin 2xPrin2. I tried this out using a published case study in which a factor correlation arose in a 2-factor DOE due to constraints (which was not a problem due to the small "standard errors" in the evaluation). The analysis of these partially correlated factors using principal component analysis, as described above, produces an almost identical result.

Gruesse,

Clothes