Publication
LIQUID: Linking Data Quality Issues to Deviations in IoT-Enhanced Process Mining
Christian Imenkamp; Yannis Bertrand; Lukas Malburg; Andrea Maldonado; Jari Peeperkorn; Pascal Schiessle; Agnes Koschmider
In: Enterprise Design, Operations, and Computing (EDOC 2026). Enterprise Distributed Object Computing (EDOC-2026), Lecture Notes in Computer Science, Springer, 2026.
Abstract
IoT-enhanced process mining allows deriving insights from raw sensor data. However, data quality issues (e.g., missing events due to inadequate sampling or range limits) are typically removed during abstraction from sensor data to event logs. This can lead to faulty decisions. Hence, the contextual information needed to explain conformance deviations is lost. This hinders analysts' ability to diagnose root causes and decreases trust in process mining results. We present LIQUID (LinkIng Data QUality Issues to Deviations), a quality-aware pipeline extension that preserves data quality metadata across transformation steps and applies probabilistic causal backtracking to trace deviations to their sensor-level origins. Evaluation results on five real-world IoT datasets show high detection accuracy and reliable root-cause attribution, complemented by actionable explanations.
