Improved Reinforcement Learning for Preventing Consumer Food Waste in Volatile Food Markets

Abstract Lack of proper planning is pointed out as one of the main culprits behind the significant consumer-end food waste. This observation can be used to build effective solutions, especially in digitalized societies. This paper proposes a reinforcement learning (RL) based decision support unit (DSU) in order to prevent food waste specifically for household and restaurant kitchens. The proposed RL model is improved with the help of mathematical modeling. The DSU assumes a generic kitchen environment for improving three important aspects of food management: the total shopping budget, the unused amount of ingredients, and the nutrition level offered by the selected menu. The RL-based algorithm is built to generate menu recommendations in a volatile market. All the practical contributions of the proposed planning tool to “monitor”, “prevent” and “reuse” levers of sustainable food strategy are summarized in a generic kitchen example. Results indicate that the proposed DSU can save around 30% from monthly shopping costs concurrently improving unused ingredients and nutritional intake. More insights about industrial processes such as package sizing and integration with “food sharing” services are also highlighted. Offering only one extra package size option for each ingredient in a market with volatile prices can help to diminish unused amounts completely and cut total shopping costs by 68% in a planned kitchen. Furthermore, the contribution of integrating mathematical optimization results into basic reinforcement learning implementation is observed as a 40% more reduction in the unused ingredients amount especially when the volatility of the market increases due to any disruption.