subselected

Subselected refers to the process in partial least squares discriminant analysis (PLS-DA) of selecting a subset of components from a wider set of components to represent the differences between different classes in a dataset. This process involves ranking the components based on their contribution to class separation and selecting those components that contribute the most to the discrimination between classes.

The objective of subselection in PLS-DA is to identify a set of components that can effectively distinguish between different classes in a dataset, reducing the dimensionality of the data and minimizing the risk of overfitting. By identifying the most contributing components, researchers can achieve better model interpretability, reduced computational complexity, and improved accuracy of the discriminant analysis model.

The subselection process involves iterating over the components, examining their contribution to class separation, and selecting those that make the most significant contribution. Common criteria used for subselection include the variance explained by each component, canonical score coefficients, and the explained variances between classes. The selected components are then used to construct the PLS-DA model.

Subselection can be performed using manual and automated methods. The former involves manual inspection and selection of components based on their contribution, while the latter employs algorithms, such as recursive feature elimination, to select the best components. Automated methods can be sensitive to the choice of parameters and, therefore, may not always yield optimal results.