Multi-elemental determination of cacao beans from participating farms in the Philippines using a handheld XRF analyzer

• Main Author: Ingalla, Dianna Joy Rivera
• Format: text
• Language: English
• Published: Animo Repository 2022
• Subjects: Cacao—Philippines—Identification; Biochemistry
• Online Access: https://animorepository.dlsu.edu.ph/etdb_chem/22; https://animorepository.dlsu.edu.ph/context/etdb_chem/article/1020/viewcontent/Multi_elemental_Determination_of_Cacao_Beans_from_Participating_F2.pdf
• Institution: De La Salle University

With the growing demand for single-origin products, especially for premium products like chocolate, the need to conduct origin tracing on key raw materials such as cacao beans has become evident. Since low production capacity is one of the main challenges faced by the cacao industry in the Philippines, cacao has become prone to fraudulent practices that threaten producers, consumers, the industry, and food regulatory authorities among others. As a way to combat such practices, it is a must to establish a fingerprint for cacao that are cultivated in the Philippines because such data are useful for traceability and origin identification. In this study, 23 samples of raw cacao were submitted by cacao producers from different regions in the Philippines which were subjected to multi-elemental determination using a handheld x-ray fluorescence (XRF) analyzer. There were 9 geographical regions involved in the study, namely Region II, Region IV, Region V, Region VII, Region IX, Region X, Region 11, Region 12, and BARMM. Across these regions, 24 elements (Mg, P, S, Cl, K, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Rb, Sr, Y, Nb, Pd, Cd, Ba, Pt, Au, and U) were detected for the cacao samples. Using the multi-elemental data, a machine learning tool, particularly random forest (RF) was trained to classify a dataset of sample measurements according to region of origin. Afterward, a fine-tuned RF model with a training accuracy of 95.65% was generated. Given by the confusion matrix of the improved model, it was found that the fine-tuned RF model was able to accurately assign most of the cacao samples in terms of region.