Gasoline classification using near infrared (NIR) spectroscopy data: comparison of multivariate techniques

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near infrared (NIR) spectroscopy is a versatile, non-destructive technique for analyzing multicomponent chemical systems.
• Applications span petroleum refining, food science, pharmaceuticals, and combustion product analysis.
• Accurate classification of complex mixtures like gasoline is crucial for quality control and process monitoring.

Purpose of the Study:

• To evaluate and compare the efficacy of nine different multivariate classification methods for gasoline classification using NIR spectra.
• To determine the optimal chemometric approach for distinguishing gasoline types and sources.
• To assess the performance of various algorithms, including K-nearest neighbor (KNN), support vector machines (SVM), and artificial neural networks (ANN-MLP).

Main Methods:

• Utilized three distinct datasets of NIR spectra (450, 415, and 345 samples) within the 14,000-8000 cm(-1) region.
• Applied and compared nine multivariate classification algorithms: LDA, QDA, RDA, SIMCA, PLS, KNN, SVM, PNN, and ANN-MLP.
• Classified gasolines into 3 or 6 categories based on source (refinery/process) and type.

Main Results:

• NIR spectroscopy demonstrated high effectiveness for gasoline classification, outperforming NMR and GC in this context.
• K-nearest neighbor (KNN), Support Vector Machines (SVM), and Probabilistic Neural Network (PNN) emerged as the most successful classification methods.
• The Artificial Neural Network-Multilayer Perceptron (ANN-MLP) approach, particularly when combined with Principal Component Analysis (PCA), yielded significantly poorer results than anticipated.

Conclusions:

• NIR spectroscopy is a powerful tool for the effective classification of gasoline.
• KNN, SVM, and PNN offer superior performance for gasoline classification compared to other tested multivariate methods.
• Further chemometric and vibrational spectroscopy research on complex systems can benefit from these findings.