Potential of Machine Learning in Demand Forecasting Based on Point of Sales Data for Food Producers - Master’s thesis

Aug 5, 2026

Upload by Ezra

This master’s thesis investigates the potential of machine learning in demand forecasting based on Point of Sales (POS) data for food producers. The research evaluates how POS data can be efficiently used to improve forecasting, comparing traditional techniques, machine learning models, and hybrid or ensemble approaches. Using a robust historical POS dataset, results show that machine learning does not outperform traditional methods in this case, yet highlights valuable feature potential. Properly applied, these features can improve forecast accuracy and reveal patterns to support inventory and supply-chain planning, while also exposing challenges such as handling outliers and integrating external factors and ML into existing infrastructures.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages88

Reading Time3 h 42 min

AccessFree to download

Abstract

Central goal and approach

Comparative evaluation of forecasting techniques

Findings and implications for forecasting

Challenges and future work

Preface

Abbreviations

Master’s thesis NT NU Norwegian University of Science and Technology Faculty of Engineering Department of Mechan ica l and Industrial Engineering Simen Emil Eide Næss Potential of Machine Learning in Demand Forecasting Based on Point of Sales Data for Food Producers Master’s thesis in Engineering & ICT Supervisor: Anita Romsdal June 2023 Simen Emil Eide Næss Potential of Machine Learning in Demand Forecasting Based on Point of Sales Data for Food Producers Master’s thesis in Engineering & ICT Supervisor: Anita Romsdal June 2023 Norwegian University of Science and Technology Faculty of Engineering Department of Mechanical and Industrial Engineering Abstract This master’s thesis investigates the ”Potential of Machine Learning in Demand Forecasting Based on Point of Sales (POS) Data for Food Producers.”The central goal was to evaluate how POS data can be efficiently utilized by food producers to improve demand forecasting. The research involved an exhaustive comparison of various forecasting techniques, including traditional methods, machine learning (ML) models, and hybrid or ensemble approaches. Using a robust dataset of historical POS data, the study scrutinized the efficacy of these methods in generating either insightful or practically useful demand predictions for food producers. Interestingly, while ML models did not exhibit superior performance compared to traditional methods in this instance, they revealed significant potential features. When appropriately implemented, these features could lead to enhanced forecast accuracy and facilitate the uncovering of intricate patterns in the data. Such insights could assist food producers in strategic decision-making related to inventory management and supply chain planning. However, the research also highlighted several challenges that need further exploration. These include managing outliers, integrating unforeseen external factors, and incorporating ML models effectively into existing supply chain infrastructures. In conclusion, the study underscores that, while ML models require further refinement for practical application in demand forecasting, they hold promise for improving forecast efficacy when harnessed appropriately. This conclusion sets the stage for further research in the field of ML-aided demand forecasting within the food production industry. Sammendrag Denne masteroppgaven undersøker ”Potensialet for maskinlæring i etterspørselsprognoser basert p˚a Point of Sales (POS) data for matprodusenter.” Hovedm˚alet var ˚aevaluere hvordan POS-data kan bli effektivt utnyttet av matprodusenter for ˚aforbedre etterspørselsprognoser. Forskningen involverte en grundig sammenligning av forskjellige prognoseteknikker, inkludert tradisjonelle metoder, maskinlæringsmodeller (ML), og hybrideller ensemblemetoder. Ved ˚a bruke en robust datasett av historiske POS-data, studerte forskningen effektiviteten av disse metodene i˚a generere enten innsiktsfulle eller praktisk nyttige etterspørselsprediksjoner for matprodusenter. Interessant nok, selv om ML-modeller ikke viste overlegen ytelse sammenlignet med tradisjonelle metoder i dette tilfellet, avslørte de betydelig potensielleegenskaper. N˚ar disse egenskapene implementeres p˚a riktig m˚ate, kan de føre til forbedret prognosenøyaktighet og lette avdekkingen av intrikate mønstre idataene. Slike innsikter kan hjelpe matprodusenter i strategiske beslutninger relatert til lagerstyring og forsyningskjedeplanlegging. Imidlertid fremhevet forskningen ogs˚a flere utfordringer som krever ytterligere utforskning. Disse inkluderer h˚andtering av outliers, integrering av uforutsette eksterne faktorer, og effektiv integrering av ML-modeller i eksisterende forsyningskjedeinfrastrukturer. Til slutt understreker studien at, mens ML-modeller krever videre raffinering for praktisk bruk i etterspørselsprognoser, holder de løfte om ˚a forbedre prognoseeffektiviteten n˚ar de blir utnyttet p˚a riktig m˚ate.

FAQs

What is the main goal of the thesis?

To evaluate how food producers can efficiently use Point of Sales (POS) data to improve demand forecasting.

Which forecasting approaches are compared in the research?

Traditional forecasting methods, machine learning (ML) models, and hybrid or ensemble approaches are compared using historical POS data.

What key challenges are highlighted for practical ML use?

Managing outliers, integrating unforeseen external factors, and effectively embedding ML models into existing supply chain infrastructures are identified as challenges.

Aug 5, 2026

Upload by Ezra