A comparison of feature selection and feature extraction techniques for condition monitoring of a hydraulic actuator

Abstract

In many applications, there are a number of data sources thatcan be collected and numerous features that can be calculatedfrom these data sources. The error of big data has lead manyto believe that the larger the data, the better the results. However,as the dimensionality of the data increases, the effectsof the curse of dimensionality become more prevalent. Further,a large feature set also increases the computational costof data collection and feature calculation. In this study, weevaluated four dimensionality reduction techniques as partof a system for condition monitoring of a hydraulic actuator.Two feature selection techniques, ReliefF and variableimportance, and two feature extraction techniques, principalcomponent analysis and autoencoders, are used to reduce theinput into three classification algorithms. We conclude thatvariable importance in conjunction with the random forest algorithmoutperforms the other dimensionality reduction techniques.Feature selection has the added advantage of beingable to remove data sources and features from the data collectionand feature calculation process that are not present inthe relevant feature subset.