Efficient and Simplified Modeling for Kerosene Processing Quality Detection Using Partial Least Squares-Discriminant Analysis Regression

Abstract

Kerosene from various refineries and crudes is used for heating and other purposes in many countries like Iraq; therefore, it is important to identify its source to recognize and tax any adulteration. In this study, a fast classification technique for kerosene marketed in Iraq was developed with the goal of identifying its quality. The samples were categorized using a supervised partial least squares discriminant analysis (PLS-DA) approach. Multivariate analyses using agglomerative hierarchal clustering and principal component analysis were utilized to identify outliers and sample dissimilarities. The dataset was divided into calibration and prediction sets. The prediction set was used to evaluate the model’s separation performance. The Q2 cross-validation was applied. The PLS-DA models achieved significant accuracy, sensitivity, and specificity, showing strong segregation ability, notably for the calibration set (100% accuracy and 1.00 sensitivity). It was found that kerosene processing can be classified rapidly and non-destructively without the need for complicated analyses, demonstrating the best results for classification even when compared with the classification outcomes of other fuels. This PLS-DA approach has never been looked at before for process quality detection, and the results are comparable to direct kerosene classification with soft independent modeling of class analogy and support vector machines.