Enhancing Retail Sustainability: Data-Driven Approach for Food Waste Detection and Prevention

Food waste represents a significant challenge within the retail supply chain, leading to considerable economic and environmental impacts. Traditional methods often fail to capture the complex, non-linear interactions between key drivers of food waste, resulting in less effective management strategies. This study aims to improve the performance of food waste detection by employing machine learning (ML) algorithms, including tree- based algorithms, neural networks, and regression models. We assess the ML models on a dataset that includes 33 temporal, meteorological, and operational features from a multi-branch grocery in a mega city. Furthermore, this research utilizes SHapley Additive exPlanations (SHAP) values for a comprehensive feature importance analysis. The results show that models such as CatBoost and ID CNN outperform traditional methods. These models provide more reliable detection and valuable insights into the factors related to food waste. We also identify the most impactful features in waste generation, providing valuable insight for retailers in sup- ply chain planning. This practical implication offers actionable feedback for improving sustainability and operational resilience in the retail supply chain.