A data-driven machine learning model for forecasting delivery positions in logistics for workforce planning

Abstract

Workforce planning in logistics is a major challenge due to increasing demands and a dynamic environment. The number of delivery positions is a key factor in determining staffing requirements. This is often predicted subjectively based on employee assessments. To improve decision making and increase both the efficiency of this important forecasting process and the use of resources in the production system, i.e. shopfloor logistics, a data-driven machine learning model with a forecasting horizon of 5 working days was developed and validated in a practical case study in a company. The results show that the novel and specifically developed model outperforms both the manual forecasting approach in practice and auto machine learning models in terms of accuracy. The outperformance is particularly strong in the short term. Based on the predicted delivery positions, an optimised workforce planning was subsequently carried out in the case study company. Limitations of the model include the fact that it was validated in only one company and that the number of picks may need to be derived for more accurate scheduling. These two aspects also represent potential for future research.