Novel Data Fusion Strategy for Second-Order Data: Multivariate Curve Resolution for the Determination of Pharmaceuticals by Means of Fluorimetric Measurements

Abstract

A new strategy is proposed for second-order data fusion based on the simultaneous modeling of two data sets using the multivariate curve resolution–alternating least-squares (MCR-ALS) model, applying a new constraint during the ALS stage, called “Proportionality of Scores”. This approach allows for the fusion of data from different sources, without requiring common dimensionality, and enables the application of specific constraints to each data set. This strategy was applied to the determination of five pharmaceutical contaminants (naproxen, danofloxacin, ofloxacin, sarafloxacin, and enoxacin) in environmental water samples, by fusing two sets of excitation–emission fluorescence matrices, measured before and after photochemical derivatization. The predictive performance of the fused model was compared to individual PARAFAC models built for each fluorescence data set, showing that data fusion significantly increases precision and accuracy, as indicated by the elliptical joint confidence region test. Data fusion allowed improvement of relative errors of prediction, from 13−32% to 8–15% in validation samples and from 25−121% to 13–20% in real samples. The advantages of data fusion were evident in both cases, particularly in instances of substantial signal overlap between analytes or the presence of uncalibrated interferents with similar profiles, as demonstrated by the superior predictive capacity achieved through the proposed strategy.