Ingredient-based Forecast of Sold Dish Portions in Campus Canteen Kitchens

Publikation: Beitrag in Buch/Konferenzbericht/Sammelband/GutachtenBeitrag in KonferenzbandBeigetragenBegutachtung

Beitragende

Abstract

In the catering industry, one major challenge is the unknown short-term demand for dish portions. Customers want to avoid queuing and desire their favorite dish according to their preferences. Meeting these demands is important for the industry but predicting future sales is a challenging task. Often, the predictions are derived manually and automated approaches are rarely applied in practice. This paper presents an ML-based forecast model using a set of derived features to predict shares and absolute numbers of dish portions per day. In particular, these features include text-based extractions of ingredients, calendar effects to model time dependencies, and favorite features to model customers' preferences. As the detailed real world evaluation shows, our approach achieves a relative model error of 15% for the prediction of dishes. Furthermore, we discuss the influence of beneficial features and assess their influence on the overall prediction quality.

Details

OriginalspracheEnglisch
TitelProceedings - 2022 IEEE 38th International Conference on Data Engineering Workshops, ICDEW 2022
Seiten111-116
Seitenumfang6
ISBN (elektronisch)9781665481045
PublikationsstatusVeröffentlicht - Mai 2022
Peer-Review-StatusJa

Externe IDs

Scopus 85134878762
Mendeley 676a4e1d-fd87-35e9-a32b-643e660348ae
dblp conf/icde/WoltmannDHL22
unpaywall 10.1109/icdew55742.2022.00023
ORCID /0000-0001-8107-2775/work/142253550

Schlagworte

Forschungsprofillinien der TU Dresden

Fächergruppen, Lehr- und Forschungsbereiche, Fachgebiete nach Destatis

Schlagwörter

  • canteen, dish, forecast, ingredients, machine learning, prediction, sales