Pre-processing and feature extraction

Feature extraction is the pattern recognition's stage in which the main signal characteristics must be distinguished from other additional or unwanted information. Also, it must be achieved keeping in mind the computing of a compact and interpretative resulting dataset from the original raw signals. Commonly, it takes place between the pre-processing and classification stages, and it frequently implies a domain change from the raw biosignals by means of mathematical transformations. However, either a feature selection stage could be required before the classification—depending on the number of obtained features or novel classification methods, such as deep-learning—can work well without the feature extraction stage. In this chapter, we focus on relevant feature extraction techniques for biosignal processing and classification, highlighting that each technique could be most suitable for a specific signal than the others. Also, they are grouped according to the nature of the domain employed for computing them as temporal, frequential, and time-frequential.