Feature selection strategies for quality screening of diesel samples by infrared spectrometry and linear discriminant analysis

Abstract

A rapid approach has been developed for the characterization of diesel quality, based on attenuated total reflectance - Fourier transform infrared (ATR-FTIR) spectrometry, which could be useful for diagnosing the sample quality condition. As a supervised technique, linear discriminant analysis (LDA) was employed to process the spectrometric data. The role of variable selection methods was also evaluated. Successive projection algorithm (SPA) and genetic algorithm (GA) feature selection techniques were applied prior to the discriminative procedure. It was aimed to compare the effect of feature selection procedures on classification capability of IR spectrometry for the diesel samples according to their quality passed or quality failed situation. Predictive capability of LDA was compared with that obtained by GA-LDA and SPA-LDA. Results showed 91.1%, 93.3% and 95.6% of accuracy for LDA, GA-LDA and SPA-LDA respectively. Thus SPA-LDA together with ATR-FTIR spectrometry was proposed as a fast screening analytical test for the evaluation of quality passed/failed situation in diesel samples.