Multivariate analysis of analytical signals to decipher relevant chemical information

Two main elements have recently characterized the research in the analytical field: on one hand the huge development of instrumental analysis in the direction of hyphenated techniques and, on the other hand, the huge development and decreasing cost of computers together with the increased capacity of computational tools. Moreover, new issues are presented to the analytical researcher by new regulations that impose to demonstrate that the whole process is under control, e.g. in the industrial/productive context, or in the life science context by the emerging need of systems biology. Thus, the role of chemometrics is more and more increasing and the toolbox of chemometrics-like methods has been progressively enriched.In particular, deciphering signal-fingerprinting of complex matrix samples requires a deeper consideration on the nature of signals feature, and it has to be taken into account that the information pertinent to the problem is mixed with many uninformative sources of variations that may affect part or the whole signal domain. These issues from the data analysis point of view are reflected in a greater complexity of the preprocessing/pretreatment and variable selection steps.The main focus of this chapter will be on feature selection methodology; after a concise review of the main recently proposed feature selection methods, the specific case of feature selection in the wavelet (WT) domain will then be considered. In particular, it will deal with illustration of our recent developed tools for WT-feature selection in regression and classification tasks. The discussion of different applications will be the core of the work, to illustrate the effectiveness of the integration of both basic (simple) and more advanced methodologies together, with a complete strategy embracing data exploration, modelling, data display-interpretation and validation.