Back to search

University of Southampton

Hybrid machine learning models for enhanced sales forecasting

Abstract

dc:description.abstract

Sales forecasts are essential to every business’s strategic plans and can both save the company money and increase its competitive advantage. However, many current businesses underestimate the opportunities which accurate forecasts provide and rely solely on judgemental forecasts from experts within the business. Machine learning and statistical forecasting methods are used by both academics and practitioners to increase the accuracy of these forecasting methods and can be further improved by combining knowledge from within the business with the statistical and machine learning techniques presented in this work. The models introduced in this study combine domain knowledge with data-driven approaches to improve forecasting on small datasets. The first approach in this work gathers global sales pipeline data to build a short-term sales forecast for a newly proposed dynamic cluster-based Markov (DCBM) model. By applying a newly developed algorithm, which first clusters the training and test set, the prediction of future sales for the next three months can be improved over a regular Markov transition model. The second proposed approach applies product lifecycle (PLC) information to improve the sales forecast. The accuracy of the sales forecast was increased for all 11 years for a luxury car manufacturer, comparing the newly developed PLC detrending approach to a common detrending by differencing approach in a seasonal autoregressive integrated moving average (SARIMA) framework. In a third model, the DCBM and PLC approaches are synthesised by using a SARIMA- long short-term memory (LSTM) framework capable of combining different data sources and thereby further increasing sales forecasting accuracy. The SARIMA-LSTM was able to predict the changes in sales<br/>occurring during the COVID-19 pandemic. All new models support short- and mid-term sales forecasting up to 12 months and represent an extension of knowledge in the area of sales forecasting.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2021

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Lechner, Albert
Advisor dc:contributor.advisor
  • Gunn, Stephen
ORCID person:identifier.orcid

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Lechner, Albert. Hybrid machine learning models for enhanced sales forecasting. doctoral thesis, University of Southampton, 2021.