A Supervised Algorithm for Water Classification by Voltammetric Measurements

Field characterisation of liquids, both for continuous monitoring and for sporadic measurements, may take advantage from the technology of taste sensing devices based on voltammetry, through the implementation of adequate sensor and signal analysis techniques. The need of designing devices suitable for field use, both for classification purposes and for change detection applications, suggests a methodological activity specifically oriented to such a target. This work particularly focuses on the dimensionality reduction approach, a crucial aspect in the view of porting the developed signal processing algorithms to embeddable field devices. The sensor device used, the methodological approach to the processing of the signal and an application experiment are described here. The proposed methodology relies on a three-step procedure, where a preliminary feature selection made on transformed coefficients of the observation vectors is followed by a further dimensionality reduction based on an iterative orthogonalization algorithm. Resulting data, represented in a reduced dimensionality space, are then fed to a classifier and displayed accordingly