Feature Selection with Sensitivity Analysis for Direct Kernel Partial Least Squares (DK-PLS)

Mark J. Embrechts
Rensselaer Polytechnic Institute, Troy, NY 12065

Partial-Least Squares (PLS) is a popular method in chemometrics for datasets with a large number of features with a high degree of collinearity. This  presentation introduces a direct kernel version of PLS. The direct kernel methodology makes the PLS model nonlinear and considers the kernel transform as a data preprocessing step.

The emphasis of the presentation will be on the importance of kernel centering and the application of successive iterative sensitivity analyses in a bootstrap mode for feature selection.