PantryPal: A Machine Learning–Driven Ensemble System for Restaurant Demand Forecasting and Waste Reduction

Food waste continues to be one of the most per- sistent challenges in the food service industry, driven largely by inaccurate demand forecasting and inefficient inventory planning. Restaurants frequently experience overproduction, stockouts, and unnecessary spoilage due to volatile customer footfall, supply- chain delays, and highly seasonal consumption patterns. Pantry- Pal is a machine learning–driven forecasting system designed to address these inefficiencies through an ensemble of boosting models, interpretable statistical models, and deep learning archi- tectures. The system integrates automated data normalization, structured feature creation, quantile-aware predictions, uncer- tainty modeling, and a production-ready retraining pipeline. Experiments demonstrate that PantryPal reduces mean absolute error (MAE) by 46% compared to classical baselines, and by 8% relative to the strongest single model. By generating reliable multi-horizon forecasts and purchase recommendations, PantryPal establishes a foundation for sustainable, cost-efficient, and scalable restaurant inventory management.