Philosophical View on Causal Knowledge

The philosophical foundations of causal knowledge offer valuable insights for advancing process mining beyond purely data-driven correlations. Building on ideas from David Hume and Judea Pearl’s causal hierarchy, causation focuses on directional relationships and domain understanding rather than simple statistical links. In business processes, domain experts naturally recognize these causal structures—for example, that a customer order leads to an invoice, not the other way around. Integrating this prior knowledge into process discovery makes models more robust to noise, reduces spurious relationships, and creates simpler, more causally consistent results. This shifts process mining from correlation-based analysis toward truly causality-aware process intelligence.

As a forefather of the philosophy of causation, Hume noted that all knowledge comes from experience and that it is based on associations between perceived events. Waldmann adopted this idea in his work on knowledge-based causal induction, indicating causal directionality as the fundamental factor for determining how statistical correlations are understood. The term causation can be further differentiated by Pearl’s three-level causal hierarchy highlighting the role of causal knowledge in helping to associate, intervene, or counterargue.

Regarding causal knowledge concerning business processes, experts with years of acquired domain-specific experience represent a valuable resource for process improvement. Experience provides process experts with a precise understanding of causal relationships between individual activities of business processes. For instance, a process owner of an order-to-cash process might readily understand that a customer order eventually leads to an invoice being created. Intuitively, it is clear to the process owner that, oppositely, an invoice followed by the customer order would contradict the causal logic of the process. In the process mining research field, most discovery algorithms do not leverage causal process knowledge. Instead, they consider data as the “single source of truths” to behaviors while overlooking domain-specific reasons.

In an experimental setting, Rembert et al. develop and test a process discovery algorithm that integrates prior knowledge. The results indicate that prior knowledge increases the robustness against noise, subsequently reducing the likelihood of measurement and ordering errors, particularly for processes with a higher degree of infrequent behavior. Similarly, Diamantini et al. exploit knowledge in complex domains with highly variable processes as a means to repair event logs and produce more realistic models. Waibel et al. use a causal template that helps process analysts integrate a causal order into discovering the process structures with a focus on control-flow.

Compared to approaches that do not integrate domain knowledge, the approach generates much simpler models with higher conformance to the defined causality by reducing the number of self-loops and spurious arcs. Lu et al. propose a semi-automated approach to detecting log patterns in process discovery, using human reasoning to evaluate, modify, and extend pattern types.

Philosophical View on Causal Knowledge

Was this article helpful?