Variable Selection for PLS-DA Models

The previous webinar discusses how and why Variable Selection can improve regression models and demonstrates several methods. These methods can also be used to help optimize Partial Least Squares Discriminant Analysis (PLS-DA) models used for classification. In this video Barry M. Wise demonstrates variable selection with VIP (Variable Influence on Prediction) and SR (Selectivity Ratio) diagnostics in the “automatic” mode on a metabolomics data set for investigating patient response to chemotherapy. The ability to use the “automatic” method on PLS-DA models will be available in the next release of PLS_Toolbox and Solo.

Try it yourself. Get a free demo of PLS_Toolbox or Solo from https://www.software.eigenvector.com/toolbox/download/  (Free account creation required)

Table of Contents

00:00 – Example: MTBL592_DSO_cleaned

04:40 – Variable selection from Cross-Validation Statistics