Decision-making in multimodal supply chains using machine learning

Purpose: To strengthen efficiency and resilience of supply chains at the same time, shippers and logistics companies needs proactive transparency about their orders. Machine Learning (ML) offers huge potential for precise predictions of complex logistics processes. This paper shows the results of a perennial research for implementing a ML-based system, which predicts multimodal supply chains, detects upcoming disruptions and provides suitable actor-specific measures. Methodology: For each process of the considered supply chain an individual prediction model is developed, using four years historical data, about 50 identified features and various ML methods. The developed cross-actor ETA was linked with preventive measures based on expert knowledge. Both was integrated into a web-based prototype of a self-learning decision support system. Findings: Thanks to the development of different ML approaches, most reliable model configurations were identified for each process. Moreover, important insights were gained regarding the availability of required data as well as the potentials and challenges of using ML-based solutions for decision-making processes in logistics. Originality: The potentials from the use of ML for predicting supply chains has only been carried out for particular processes. An integrated approach including different processes like rail transports and transshipment points as well as a linkage with prediction-based measures is still missing.