Answers are grounded in this paper and cite their sources.
Data streams from Internet of Things (IoT) devices containing sensors and actuators provide new insights into their interactions, context, and process executions in the physical world. These new data sources may open up novel opportunities to apply Business Process Management (BPM) technologies to analyze process and activity executions using established process mining techniques. However, the rather low abstraction level of data emitted from the IoT devices is often not suitable to directly apply process mining, which requires additional steps of event abstraction. Related approaches train expensive supervised machine learning models on historical sensor data to realize this event abstraction enabling only a post-mortem classification of activity executions. In this work we propose a framework to automatically generate activity detection services from IoT data with minimal human involvement to implement the event abstraction. Along with the framework, we present a software architecture focused on a flexible and extensible complex event processing (CEP) platform that achieves high-performance activity detection from IoT data streams at runtime–enabling online process analytics. Evaluations of our proof-of-concept implementation to monitor processes executed in smart manufacturing and smart healthcare show acceptable results when detecting activities that are affected by no to only small variations in the underlying IoT data. We identify several ways to improve the robustness of the activity detections regarding variations in IoT data as starting points for future work.
No takes yet. Share an insight, caveat, or question.
Seiger et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: