Feature Selection

Returning briefly to the subject of dimensionality reduction, the two main techniques to reduce data dimensionality are feature extraction and feature selection. The aim of feature selection in machine condition monitoring is to select a subset of features that are able to discriminate between instances that belong to different classes. Based on the availability of class label information, feature-selection methods can be categorised into three main groups: supervised, semi-supervised, and unsupervised. Moreover, depending on their relationship with learning algorithms, feature-selection techniques can be further grouped into filter models, wrapper models, and embedded models. The chapter presents generally applicable methods that can be used to select the most important features to effectively represent the original features. These include various algorithms of feature ranking, sequential selection algorithms, heuristic-based selection algorithms, and embedded model–based feature-selection algorithms.