Data Pre-Processing for Feature Extraction, Dimension Reduction and Concept Formation

Abstract

This chapter describes data pre-processing for feature extraction, dimension reduction, noise removal and concept formation from monitored process measurements. The discussion is concerned with capturing the features in dynamic trend signals. A dynamic trend representation is the visualisation of the changing trajectory of a variable over time and consists of many sample values. However, in order to make effective use of trends in a computer based system, it is necessary to compress the data to fewer values. One of the earliest examples of dealing with such trends is based on a real time expert system G2 [76] which uses qualitative expressions such as temperature increase or decrease as descriptors. Later, various other approaches were developed including episodes [77, 78, 79], neural networks [80] and more recently wavelets [81, 82] and principal component analysis [83]. This chapter introduces principal component analysis, wavelet analysis as well as episode representations.