Chemometric methods for classification and feature selection

Classification methods, i.e., the chemometric strategies for predicting a qualitative response, find many applications in the omic sciences, where often data are collected in order to categorize individuals (e.g., according to whether they were treated or administered a placebo or, for instance, depending on if they were healthy or ill). After a brief discussion of the differences between discriminant and modeling approaches, some of the techniques most commonly used in the omic fields are illustrated in greater detail. A part of the chapter is then devoted to illustrating the strategies for identifying the most relevant features in model building through variable selection approaches, and their role in putative biomarker identification. Lastly, the importance of validation is also addressed and a brief guideline of the available strategies is presented.