Rapid assessment and prediction of microbiological quality of raw milk using machine learning based on RGB-colourimetric resazurin assay

Abstract

The feasibility of adapting an RGB-based colour sensor to determine raw milk quality based on the resazurin test was demonstrated and a model for rapid milk quality prediction proposed. One hundred and two raw milk samples containing different microbial concentrations were subject to a resazurin assay in a six-well microplate configuration and then to the colourimetric device to measure the resazurin test colours as red, green, and blue. A dataset of RGB colour measurements and corresponding microbial concentrations was created, from which a machine learning model for milk quality prediction was developed. A support vector machine model that was considered most suitable for this purpose demonstrated 100% prediction accuracy for milk with acceptable or “low” microbial concentrations (<5.0 × 105 cfu mL−1) and 96% accuracy for milk with unacceptable or “high” microbial concentrations (>1.0 × 107 cfu mL−1) but was less accurate for milk in the intermediate class.