| Spectroscopic data paired with chemometric modeling methods has become a powerful, cost effective, and rapid analytical tool over the last decade. However, large global spectral libraries spanning numerous sample matrix differences and instrument conditions are often nonlinear in relation to the measured chemical prediction property of interest. These differences result in lower model prediction accuracies. One solution to overcoming nonlinear relationships is to use local modeling techniques. In local modeling, a unique subset of calibration samples are selected from a global library for each specific target sample. Many local modeling algorithms rely on one or two spectral similarity measures for selecting calibration samples while overlooking similarities based on chemical properties. Current local modeling methods also require predetermined selections of specific variables including similarity merits, number of samples, and regression model tuning parameters. This work explores techniques for selecting local calibration samples that are both spectrally and chemically similar to the target sample while reducing the number of predetermined variables required. The process of local adaptive fusion regression (LAFR) employs many unique aspects, including data fusion and cross modeling, to select matrix matched calibration samples. Aspects of the local adaptive fusion regression process are first used to demonstrate why data fusion and cross modeling techniques are successful for identifying matrix matched calibration sets. The automated LAFR process, using these same techniques, then demonstrates how matrix matched local calibration sets are consistently formed and selected. |