PLSLDA

Overview

PLSLDA combines PLS Regression and Linear Discriminant Analysis. It can handle multiclass problem i.e. the target variable can have \(K \left(K \geq 2\right)\) classes. It relies on the same principle (as CPLS and PLSDA) about the number of components detection.

Description of the method

There are two main steps in the learning process. Firstly, using the same coding scheme as PLSDA, we launch the PLS algorithm. Secondly, we launch the linear discriminant analysis on the X component scores (factors). Because these factors are orthogonal, the DISCRIM is more reliable. This kind of data transformation is very useful when the original input variables are highly correlated.

Predictive idea

The classification rule is the same as the PLSDA component.

Note

Because we combine two approaches, it is very difficult to evaluate the influence of the original features. For this reason no information is provided about the relevance of features, but about the pls components.

Number of components

In PLSLDA procedure, we can explicitly specify the number of components, with the parameter n_components, for NIPALS [1] algorithms.

References

[1]

NonLinear Iterative Partial Least Squares.