Partial Least Squares Discriminant Analysis (PLS-DA) applied to smectites classification
PLS-DA is a machine learning tool that combines PLS regression and PCA. Through PLS regression a model describing the relationship between a categorical variable and several numerical variables is provided and this model is used for the dimensionality reduction with a PCA, in such a way that the underlying structure of the data is described and the discrimination between the categories is maximized. PLS-DA involves several steps, cross-validation is an important step in using PLS-DA as a feature selector, classifier or even just for visualization.
Smectites are important industrial minerals with great variability in their chemical composition. Their crystallo-chemistry of major elements is well known, but it is not known if trace elements are related to the structure. PLS-DA allows us to find the set of elements that best discriminate between two crystallo-chemical categories of smectites (dioctahedral or trioctahedral) to conclude certain trace elements are related to their structure.
Palabras clave: pls regression discriminant analysis dimensionality reduction feature selection smectites