Potential for using guest attendance forecasting in Swedish public catering to reduce overcatering

Christopher Malefors,Per-Anders Hansson,Ingrid Strid,Mattias Eriksson

doi:10.1016/j.spc.2020.08.008

Christopher Malefors, Per-Anders Hansson + Show 2 more

Open Access

https://doi.org/10.1016/j.spc.2020.08.008

Copy DOI

Abstract

Food waste is a significant problem within public catering establishments, caused mainly by serving waste arising from overcatering. Overcatering means that public catering establishments rarely run out of food but surplus ends up as food waste. The challenge is to find a solution that minimizes food waste while ensuring that sufficient food can be provided. A key element in this balancing act is to forecast accurately the number of meals needed and cook that amount. This study examined conventional forecasting methods (last-value forecasting, moving-average models) and more complex models (prophet model, neural network model) and calculated associated margins for all models. The best-performing model for each catering establishment was then used to evaluate the optimal number of portions based on stochastic inventory theory. Data used in the forecasting models are number of portions registered at 21 schools in the period 2010–2019. The past year was used for testing the models against real observations. The current business as usual scenario results in a mean average percentage error of 20–40%, whereas the best forecasting case around 2–3%. Irrespective of forecasting method, meal planning needed some safety margin in place for days when demand exceeded the forecast level. Conventional forecasting methods were simple to use and provided the best results in seven cases, but the neural network model performed best for 11 out of 21 kitchens studied. Forecasting can be one option on the road to achieve a more sustainable public catering sector.

Highlights

The global population is estimated to reach 9.6 billion people around 2050 (United Nations, 2019)
The data used for the analysis comprised material from 21 public school kitchens in Sweden, which were selected as suitable test subjects for forecasting models because they were willing to share their data upon request and had data available for several years
There was sometimes very little differences between the models, as indicated by the results in Table 2, where the moving average model with a five-day window was good as its moving average equivalent with a two-day window or the neural network model in two of the cases (Kitchen 3 and 8)

Summary

Introduction

The global population is estimated to reach 9.6 billion people around 2050 (United Nations, 2019). Feeding a growing population is a challenge that needs urgent attention, since agricultural production is a significant driver for transgression of several planetary boundaries and poses a threat to boundaries currently regarded as lying in the safe zone (Campbell et al, 2017). Acute interventions are needed at global scale to achieve a sustainable food system that can deliver on-point. One of many interventions to create a sustainable food system is to target the vast amounts of food that are currently destroyed, spoiled, or dumped for various reasons, and reduce the level of food waste by 75% by 2050 (Springmann et al, 2018). The Food and Agriculture Organization of the United Nations (FAO) suggests that more effort is needed to map the food waste situation and identify how food waste reduction on an overarching level can be achieved (FAO, 2019). Primary data and methods to battle food waste are lacking, and improvements are badly needed in most cases (Xue et al, 2017)

Objectives

Methods

Results

Discussion

Conclusion