A wavenumber selection approach for sample classification in the petroleum sector

Abstract

In recent years, spectroscopy techniques such as Near-infrared (NIR) and Fourier Transform Infrared (FTIR) have been adopted as analytical tools in different fields. A spectrum of a sample usually has hundreds of wavenumbers, fact that can jeopardize the accuracy of statistical analysis, being the variable selection an important step in prediction and classification tasks based on spectroscopy data. This paper proposes a novel methodology for wavenumber selection in classification tasks, applied in two data sets from the petroleum sector. The method consists of two main stages: determination of intervals based on the distance between the average spectra of the classes and the selection of the most suitable intervals through cross-validation. An improvement of 11.52% in the misclassification rate was achieved for a NIR spectra data set of diesel, decreasing from 11.71% to 10.36% after the application of the proposed method. For a biodiesel FTIR data set the method proved to be robust, achieving a zero misclassification rate after the selection process, compared to its initial value of 4.71%.