Using Reinforcement Learning to Reduce the Waste of Freshly Made Food While Maintaining Student Satisfaction in the Dining Hall
Food waste in higher education institutions amounts to a staggering 22 million pounds annually, according to the Food Recovery Network. While prior research has explored various strategies to combat this issue, few have addressed the operational aspect of dining halls without trying to predict or control student behavior. This research proposes the application of Reinforcement Learning (RL) to efficiently manage the preparation of freshly made food in dining halls, minimizing food waste while still meeting students' needs. The study begins by simulating the process of serving sandwiches in a dining hall. Monte Carlo methods and Q-learning are used to train a policy that helps determine the optimal number of sandwiches employees should prepare based on the current state. Comparing these RL methods with a baseline policy reveals that the Monte Carlo method performs most effectively. This demonstrates the potential of reinforcement learning techniques in reducing food waste in dining halls by optimizing food preparation decisions. Beyond its application in campus dining, this study has broader implications for companies dealing with perishable products, offering insights into inventory planning. By integrating RL strategies into food service operations, we can enhance resource allocation, reduce environmental waste, and provide valuable lessons for the food industry, thereby addressing a critical sustainability challenge. In the future, I could increase the complexity of the simulation by implementing a Last-In-First-Out (LIFO) policy to model how students select sandwiches or increasing the variety of the sandwiches offered, allowing for a more realistic assessment of food waste reduction strategies.