Comparative study of classic and fuzzy time series models for direct materials demand forecasting
Zakrytnoy, Sergey (2021)
Diplomityö
2021
School of Engineering Science, Tuotantotalous
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2021050428645
https://urn.fi/URN:NBN:fi-fe2021050428645
Tiivistelmä
In many industries, direct materials budgeting is an essential part of financial planning processes. In practice, it implies predicting quantities and prices of dozens and hundreds of thousands of different materials that will be purchased by an industrial enterprise in the upcoming fiscal period. Lack of collaborative processes over the length of the supply chain, distortion effects in demand projections and overall uncertainty cause the enterprises to rely on internal data to build their budgets.
This research addresses the need for a scalable solution that would use mathematical models to reveal intrinsic patterns in historical purchase quantities of direct materials and generate automatic forecast suggestions. Business context and limitations are explored, and relevant time series forecasting methods are shortlisted based on existing practice described in academic research. Furthermore, anonymized datasets of direct materials purchases from three industry partners are used to evaluate predictive performance of the shortlisted methods. Quantitative part of the study reports an improvement in prediction accuracy of up to 47% compared to the currently used naïve approach, with fuzzy time series models being most appropriate for the intermittent time series in question.
By means of a comparative study, the research demonstrates that it is feasible to apply univariate models in direct materials budgeting processes, and suggests further topics such as implementation complexity that need to be explored prior to taking those models into use.
This research addresses the need for a scalable solution that would use mathematical models to reveal intrinsic patterns in historical purchase quantities of direct materials and generate automatic forecast suggestions. Business context and limitations are explored, and relevant time series forecasting methods are shortlisted based on existing practice described in academic research. Furthermore, anonymized datasets of direct materials purchases from three industry partners are used to evaluate predictive performance of the shortlisted methods. Quantitative part of the study reports an improvement in prediction accuracy of up to 47% compared to the currently used naïve approach, with fuzzy time series models being most appropriate for the intermittent time series in question.
By means of a comparative study, the research demonstrates that it is feasible to apply univariate models in direct materials budgeting processes, and suggests further topics such as implementation complexity that need to be explored prior to taking those models into use.