Literature review and process development for inbound transport demand forecasts

Order dates and delivery times in the inbound-logistic are subject to extern influences. The resulting fluctuations are imposing a challenge for the planning of the necessary transport and handling capacities. In order to plan in an environment of uncertain demands, the method of demand forecasting provides an approach to improve the planning accuracy. Due to the high volatility of the underlying demand time series machine learning (ML) based methods are identified to enable time series predictions of transport demands. For the identification of suitable ML methods, a representative and structured literature review is conducted. Following the analyses of the review an approach for the implementation of ML models in a data mining use case for transport demand predictions is presented. To ensure the actual implementation of the identified Long-short-term-memory (LSTM) model a process model based on the Cross-Industry-Standard-Process for Data mining (CRISP-DM) is formulated to support the implementation with specific steps required for this use case. With this approach, a consistent concept is created which enables the selection and implementation of a time series forecast for transport demands and thus the optimization of transport demand planning.