Development of a Knowledge-Based Framework for Demand Management for Refrigerated and Shelf-life Constrained Food Supply Chains

Fresh produce like fruits and vegetables and food products requiring refrigeration like milk, are characterised by aspects such as perishability, short shelf-life, and high demand fluctuation, which make demand forecasting a vital process affecting not only business profits, but also the amount of waste and the level of customer satisfaction.

The research aim is to investigate the feasibility of the knowledge engineering approach (KEA) as an alternative to statistical data analysis to improve demand forecasting, specifically for small and medium-sized enterprises (SMEs) in the fresh food supply chains (FSCs).

The methodology comprised statistical data analysis techniques ranging from simple (correlation analysis) to more advanced (support vector machines) implemented using demand data for specific products provided by a fruit and vegetables wholesaler, to find the most influential internal and external factors affecting demand. After evaluating the results from these techniques, KEA was explored as a possible option to improve demand forecasting. For knowledge acquisition, a questionnaire about demand management was developed, which was used during structured interviews as a tool to externalise the tacit knowledge that the experts use to support decision making for daily demand forecasting. Further sources of information used were on-site interviews and a knowledge engineering tool (KET), updated regularly by the demand planner to aid identifying reasons behind differences between demand prediction and actual orders.

The results from the statistical data analysis and support vector machines experiments showed no significant improvement in the accuracy of prediction. The outcomes from the knowledge acquisition process showed the high level of difficulty involved in externalising tacit knowledge from experts. However, the KEA provided direct insight from the experts about the issues affecting demand forecasting in the FSCs. Moreover, the knowledge gathered through the on-site interviews and the KET was used to define rules that could be used to further develop a knowledge-based framework to support demand forecasting. The proposed framework suggests using the KEA alongside other prediction methods to improve prediction accuracy and highlights the importance of targeting the right experts, in order to aid generalising expert knowledge when defining useful rules to support decision making.