A chemometric strategy based on peak parameters to resolve overlapped electrochemical signals

A new chemometric methodology based on the use of peak parameters as direct input data into different multivariate calibration methods is proposed. Different regression techniques such as multilinear regression (MLR), partial least square regression (PLS), principal component regression (PCR) and artificial neural networks (ANN), were utilized in order to resolve hard overlapped electrochemical signals belonging to the well-known Tl+/Pb2+ system, which was used as a benchmark. This strategy was studied as an alternative to traditional procedures that apply pre-treatment techniques (dimension reduction methods or feature selection processes) to the full voltammograms of the signals. Good predictive and effective models were obtained, being the RMS errors very similar in all cases, independent of the calibration method. However, ANN-based regression models performed slightly better. The average relative errors ranged from 5 to 10% for Tl+ and from 4 to 12% for Pb2+. A study of the relevance of the voltammetric peak parameters was also carried out. This parameters-based strategy can involve a fast and efficient alternative to resolve multicomponent systems in those analytical techniques whose signals can be represented by peak parameters.