Standardization of NIR data to identify adulteration in ethanol fuel

This work proposes the use of classification transfer based on standardization methods to investigate adulteration in hydrated ethyl alcohol fuel (HEAF) samples analyzed in two different NIR spectrometers. For this purpose, direct standardization (DS) and partial direct standardization (PDS) methods were applied to a transfer sample set (measured in the two devices) in order to estimate an adjustment function between the data obtained from two different instruments. For purposes of comparison, classification models based on linear discriminant analysis (LDA) with previous selection of variables by successive projections algorithm (SPA), and partial least squares-discriminant analysis (PLS-DA) were developed before and after the standardization procedure. The performance of classification models was evaluated according to correct classification rate (CCR) for the prediction set. Before the standardization procedure, a low correct classification rate of 50% was found for both SPA-LDA and PLS-DA models when samples with spectra were recorded in a different instrument than that used to build classification models. However, after standardization, we observed a significant increase in correct classification rate for these samples, reaching 97.5% and 100% for SPA-LDA and PLS-DA model respectively.