Logistics processes depend heavily on changing conditions and making accurate forecasts is becoming more and more important to preventing delays and/or predict risks. Predictive Process Monitoring has advanced through deep learning and process-mining approaches, yet current methods often lack interpretability and lose accuracy when context varies. Recent research shows that contextual factors can improve predictions, but their integration into transparent, model-driven frameworks remains limited. This article presents a context-aware predictive approach that filters historical event logs by important attributes, discovers process models with the Inductive Miner, and predicts future activities and timestamps using token-based replay and polynomial regression. Experiments with real logistics data show that incorporating context reduces prediction errors, while the use of process mining ensures an interpretable and operationally practical forecasting solution for logistics environments.
Barbeiro et al. (Thu,) studied this question.