Preventing waste in food supply networks - a platform architecture for AI-driven forecasting based on heterogeneous big data

Abstract

Waste is a major challenge in today´s food supply networks. It induces economic, ecologic, and social implications. Current advanced supply and demand planning frameworks consider waste as an expected lost value in a profit-seeking trade-off. Consequently, existing initiatives for food re-distribution tackle the problem at the end of the supply chain, i.e., at the retailer stage. Advanced forecasting is foreseen as a facilitator of planning towards food waste prevention. However, companies in food production and retail seldomly utilize advanced forecasting methods, i.e., artificial intelligence. Deterrents for inadaptability rank from the differentiation between the role of pure data driven reliance and the human factor. Furthermore, no remarkable endeavor has so far combined heterogeneous external data integration with advanced forecasting techniques for food waste reduction. This paper proposes an architecture for a collaborative digital demand and supply matching platform for perishable food supply chains. Focus lies in the integration of heterogeneous big data (i.e., mobile transaction data and weather data) with AI-based forecasting methods. Building on the Design Science Research methodology, problems and requirements from the perishable food industry form the base for the platform architecture as the core artifact. The validation of the proposed architecture is undertaken by a panel of selected business experts from food retail and wholesale. The designed platform architecture consists of the core modules “data layer”, “logic layer”, “interaction layer”, and “data visualization”, and includes governance rules for sensible knowledge protection and data security. The present work contributes to the research area of food waste prevention by combining collaborative supply chain management approaches with advanced data-driven technologies including artificial intelligence and heterogeneous big data integration. Further research will extend this study by a use-case implementation.