Detection of Fruit Pulp Adulteration Using Multivariate Analysis: Comparison of NIR, MIR and Data Fusion Performance

Abstract

This study aimed at developing control charts and classification models to investigate sugar and water addition in guava pulp applying near- and mid-infrared (NIR and MIR) spectroscopies and low-level data fusion to compare performance of them. The pulp was produced in a pilot plant (authentic samples) during the harvest season in Sao Paulo (Brazil), and part of samples was adulterated with sugar or water. Authentic and adulterated samples were analyzed by NIR and MIR. The spectra data obtained were preprocessed, and the principal component analysis was applied. MIR spectra data presents a fingerprint region, which is an important tool to differ authentic and adulterated samples. Control charts and classification models (SIMCA, k-NN, and PLS-DA), which were authenticated by external validation, were used to discriminate authentic from adulterated samples (sugar or water in different concentrations). It was possible to differentiate adulterated from authentic samples through control charts, except for water-adulterated samples using NIR spectral. The models presented excellent values of sensitivity, specificity, accuracy, and efficiency. However, k-NN presented better performance. The results obtained by data fusion presented worse performance than the models based in only one of the techniques. Therefore, these results suggested that NIR and MIR techniques can be used for adulteration detection; however, MIR control charts and k-NN models are more effective to detect sugar or water adulteration in guava pulp.