Process mining extracts actionable insights from event data recorded by information systems Aa16; ARM24. Building on these foundations, Predictive Process Monitoring (PPM) uses historical logs to forecast future process states from partial traces, including next activities, outcomes, or remaining time AIM+25; GPR+25. Most existing approaches assume a centralized data source and full process observability. This assumption increasingly breaks in cyber-physical systems (CPS), where event data is fragmented across autonomous components and constrained by privacy requirements, organizational boundaries, communication limits, heterogeneous schemas, and partial observability Br25; Li23. Distributed and federated process mining address this challenge Aa21; An25; LGT+25. However, local data in CPS is typically Non-Independent and Identically Distributed (non-IID), adding further challenges: In CPS, each component produces a diverse set of activities. Depending on the current observed cases, the activity sets between two different components may be the same, overlapping, or very different. As a result, a single global model trained on global data may dilute activity-specific regularities, whereas purely local models may lack global context. This motivates the question: how can we devise accurate predictive models when event data must remain decentralized?
Tran et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: