Dimensionality reduction

The analysis of biological signals requires very effective predictive models that take as input pre-processed signals and return as output a prediction associated with the task at hand. Whereas the predictive model is critical for the success of the model, an almost similarly important task has to do with processing inputs to reduce their initial dimensionality and identify those variables that are more relevant for building a predictive model. This is precisely the topic covered in this chapter that deals with the dimensionality reduction problem; we formulate the task in the context of supervised learning and review the two main approaches for solving it: feature selection and feature transformation. These interrelated tasks aim at processing the input signals to extract meaningful information while maintaining it at a manageable dimension. Both approaches are described, including their associated challenges and representative methodologies. The goal of this chapter is to serve as an introductory reference for non-experts in machine learning.