Reducing Food Waste in Supermarkets with Machine Learning

Around one-third of produced food is wasted, which is almost 1.3 billion tons of food per year, affecting the world economically, environmentally, and socially. Over the years, different solutions for food waste have been developed; however, machine learning within this area has not been sufficiently explored. The food rescue process is very time sensitive since the food can be close to turning rancid, so forecasting the supply of food to be donated could significantly increase the amount of food being rescued. Our objective is to optimise the use of food and reduce food waste in supermarkets. We propose a framework to predict the amount of expiring products donated with machine learning models. In the paper, a first test of the framework, using simulated data, is presented, showing that the Support Vector Regression would be the most appropriate machine learning model for the framework, giving a score of 0.732 While the paper shows that the framework can generate valuable results, the difficulties with applying it to the real world lie in the data acquisition phase. Supermarkets are private companies that might not be willing to share their business data, which is a problem for future research.