Um método baseado em aprendizado de máquina para previsão da produção de refeições em restaurantes universitários: A method based on machine learning to predict meal production in university restaurants

Food waste from university restaurants is a common problem, generating a significant loss of food produced. Estimating the amount of food to be produced is a major challenge due to the behavior of students who frequent restaurants and the number of variables that must be observed. In order to make such estimates considering these variables and this behavior, methods based on Machine Learning (ML) can be explored for this purpose. Three different ML algorithms were tested and evaluated in this context: K-Nearest Neighbors, Random Forest, and Artificial Neural Networks using data from the restaurant of the Universidade Federal de Uberlândia. This study performed an analysis and pre-processing of the data, applying them to the ML algorithms, which were evaluated and compared with the human prediction and an algorithm that uses the average as an estimator. The results showed that in three of the four scenarios, the ML algorithms did better than the human prediction or an algorithm that does not use ML. In additional comparison, the human prediction did better than one of the four scenarios. We concluded that ML might be a reasonable solution in the future to reduce waste in college restaurants.