Machine learning for predicting industrial performance: Example of the dry matter content of emmental-type cheese

Abstract

Controlling the dry matter content of cheese is essential to defining the performance of cheese production. For Emmental-type cheese, dry matter content has to be above but as close as possible to a minimal value that is defined by legislation. The means for achieving the target dry matter content was mostly left to the discretion of the cheese experts, who target a dry matter objective based on his expert knowledge and the deviation of cheese production. To date, the prediction of performance indicators, such as cheese dry matter content, can help cheesemakers to improve their production performance. Several Machine Learning models and classical statistical methods were compared to predict the dry matter of Emmental cheese for a set of data coming from one selected cheese industry. The Random Forest method emerged as the most effective model (RMSE = 0.28 and R² = 0.67). The weight of variables in explaining the variability of cheese dry matter content was also calculated, helping cheese experts to interpret the model and apply corrective actions to improve cheese production performance. The ability to predict cheese dry matter content and understand its variability from cheese manufacturing data offer new perspectives for the cheese industry. This method can be transferred to other indicators and assist in decision-making to enhance industry performance.