Forecasting future orderline quantities for third party logistics provider of warehousing operations with statistical and machine learning tools: Improving resource planning and workforce utilization
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Authors
Date
Department
Major/Subject
Mcode
Degree programme
Language
en
Pages
73
Series
Abstract
The business of offering third party warehousing services is a labour-intensive business. For this reason, resourcing is a vital task to succeed for the service provider. This is since employees cause significant part of the variable costs that service offering logistics company can control at short range. In addition, resourcing impacts directly on the operational metrics like picking process efficiency. Specially picking process causes significant costs for the service provider totalling more than 50 % of variable costs. For this reason, it is critical to predict the future volumes at decent accuracy so that resourcing could be implemented as accurate as possible without causing excess costs. Forecasting future volumes can be implemented with variety of ways from qualitative subjective evaluation of future orderlines that is often used in described environment. The qualitative forecasting if often based on management’s experience making companies dependent on their expertise. In this thesis, the quantitative approach is taken, and forecasting is implemented by traditional methods like moving averages and simple exponential smoothing, but in addition more modern approaches are used like machine and deep learning. One major distinction between these approaches is than machine learning methods don’t take any parameters beforehand but those adjust models and parameters to match the training data. Thus, the machine learning and other modern methods are more intensive computationally. The results indicate in forecasting accuracy metrics that the machine learning and deep learning methods would perform better than the traditional methods. Especially long-shot term memory seems to be performing at the best level. When looking for statistical significance for better forecasting accuracy there does not seem to be consistency in the results. Thus, it is hard to say that one method would outperform the other methods. In a nutshell, forecasting methods provide additional information for the decision makers and the planner’s and tackle down one of the major downsides that qualitative forecasts have: bias. However, no matter how good the forecasts were, the tools still are decision support tools that need a human to make the final call on resourcing.Description
Thesis advisor
Malo, PekkaKorhonen, Atte