Recent Advances in High-Level Fusion Methods to Classify Multiple Analytical Chemical Data

Abstract High-level data fusion strategies combine and integrate predictions obtained by means of individual models calibrated on single information sources. With respect to other fusion techniques, high-level methods aim at improving prediction performances, as well as reducing the total uncertainty associated with the final combined outcome. In fact, when only partial or even conflictual information is provided by the different sources, the high-level approach integrates the different information and therefore increases the reliability of the combined (fused) prediction. In this chapter, basic (i.e., majority voting) and advanced (i.e., Bayesian consensus with discrete probability distributions and Dempster-Shafter theory of evidence) high-level strategies are evaluated on real analytical multiblock datasets and their advantages and drawbacks described.