Predicting total solids in a dairy industry by applying machine learning techniques

Abstract

This paper details the experience of a technological project carried out between a major dairy industry in the province of Santa Fe and the Information Management Laboratory of the National University of Rafaela. The project involved the analysis of total solids in raw milk and was carried out using a quantitative methodology based on the CRISP-DM model. Meetings were held between the parties for the data comprehension stage. During the analysis phase, the variables to be used were determined and processed in statistical models. During the modeling process, different alternatives were analyzed using machine learning algorithms, determining that linear regression worked best. The mean error was used as a reference to evaluate these algorithms. Finally, a tool was developed, using Python programming code, adapted to the company and capable of predicting total solids. The project positioned the University as a benchmark in technologies and process improvement, as well as bringing the company closer to data science and helping it make agile and informed decisions by reducing operational times in recipe updates.