Modified locally weighted-Partial least squares regression improving clinical predictions from infrared spectra of human serum samples

► Application of LW-PLSR for the prediction of major and minor components in serum. ► Sample selection for optimum local models is strongly affected by the matrix. ► RMSEP decrease when analyte-specific regions are used for sample selection. ► However, RMSEP also decrease when the whole spectrum was used for PLS modeling. ► Combining both regions, LW-PLSR improved RMSEP values of PLS between 5 and 30%.