Classification of raw cow milk using information fusion framework

• Published in: Journal of food measurement & characterization 2021-12, Vol.15 (6), p.5113-5130
• Main Authors: Polat, Olcay, Akçok, Seda Gökçe, Akbay, Mehmet Anıl, Topaloğlu, Duygu, Arslan, Seher, Kalayci, Can Berk
• Format: Article
• Language: English
• Subjects: Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Classification; Cow's milk; Data integration; Decision making; Engineering; Farms; Food Science; Milk; Multiple criterion; Original Paper; Parameter sensitivity; Sensitivity analysis; Statistical analysis; Statistical methods; Supply chains
• ISSN: 2193-4126; 2193-4134
• DOI: 10.1007/s11694-021-01076-5

Classification of the different samples of raw cow milk based on quality is of great importance in controlling the whole process of the dairy supply chain and ensuring end-product variety. It is very difficult to determine a standard classification approach, as milk can take on different physical and chemical structures, depending on changing external conditions, due to its nature. A comprehensive approach considering multiple criteria for classifying milk into certain categories has not been presented in the literature yet. This study aims to classify raw milk according to quality by considering the multi-criteria consolidated from multiple sources. Multi-criteria decision-making (MCDM) methods assist decision-makers where there are multiple and conflicting criteria measured in different units. To classify raw milk, an MCDM based information fusion framework was used to integrate expert knowledge and instrumental data. In this framework, the AHP method was utilized to calculate the importance levels of quality criteria, and the VIKORSORT method was used to classify alternatives to quality groups. Ultrasonic sensors were used to recognize nine specified criteria and a semi-automated flow cytometer was used to measure one criterion. A real case investigation was carried out, with 87 samples taken from different farms being used to validate the proposed framework. The samples were divided into 2, 3, or 4 milk quality groups when the steps of the framework were completed. Statistical validation tests show that pH, sH, and somatic cell count were very effective in determining milk quality groups and 89.7% of samples were predicted as in correct group. Indeed, the reaction of the model to the chance of methodological parameters was showed in detail with sensitivity analysis. The findings of the study show that the framework could be used to categorize raw milk into several classes by dairy companies or institutions considering relatively critical criteria.