[Paper Review] Learning Hybrid Process Models From Events: Process Discovery Without Faking Confidence

Wil M. P. van der Aalst, Riccardo De Masellis+2 more|arXiv (Cornell University)|Mar 17, 2017

Business Process Modeling and Analysis10 citations

TL;DR

This paper proposes a hybrid process mining approach that combines formal semantics with intentional vagueness by using hybrid Petri nets to model clear causalities precisely while representing uncertain or complex dependencies informally. The method discovers formal places for strong causal relations and uses 'sure' or 'unsure' arcs for weaker or ambiguous ones, achieving scalability and interpretability without faking confidence in uncertain behaviors.

ABSTRACT

Process discovery techniques return process models that are either formal (precisely describing the possible behaviors) or informal (merely a "picture" not allowing for any form of formal reasoning). Formal models are able to classify traces (i.e., sequences of events) as fitting or non-fitting. Most process mining approaches described in the literature produce such models. This is in stark contrast with the over 25 available commercial process mining tools that only discover informal process models that remain deliberately vague on the precise set of possible traces. There are two main reasons why vendors resort to such models: scalability and simplicity. In this paper, we propose to combine the best of both worlds: discovering hybrid process models that have formal and informal elements. As a proof of concept we present a discovery technique based on hybrid Petri nets. These models allow for formal reasoning, but also reveal information that cannot be captured in mainstream formal models. A novel discovery algorithm returning hybrid Petri nets has been implemented in ProM and has been applied to several real-life event logs. The results clearly demonstrate the advantages of remaining "vague" when there is not enough "evidence" in the data or standard modeling constructs do not "fit". Moreover, the approach is scalable enough to be incorporated in industrial-strength process mining tools.

Motivation & Objective

• Address the gap between formal, precise process models and informal, scalable models used in commercial tools.
• Overcome the limitations of existing formal models that force binary classification of traces, even when data lacks sufficient evidence.
• Provide a scalable and interpretable process discovery technique that preserves uncertainty and avoids misleading precision.
• Enable formal reasoning on well-structured parts of a process while allowing informal representation for complex or ambiguous dependencies.
• Develop a method that supports both industrial scalability and reliable model interpretation for process mining applications.

Proposed method

• Construct a causal graph from event logs using configurable thresholds to detect potential causal relations between activities.
• Apply explicit quality criteria to identify and add formal places in the Petri net that represent unambiguous causal dependencies.
• Represent uncertain or weakly supported dependencies as 'sure' or 'unsure' arcs, preserving vagueness where justified by data.
• Use hybrid Petri nets as the underlying formalism, allowing both precise semantics for places and informal annotations for arcs.
• Implement the algorithm in ProM and evaluate it on real-life event logs to assess scalability and model quality.
• Leverage existing performance and compliance analysis techniques on formal parts (places), while extending them to approximate indicators for informal arcs.

Experimental results

Research questions

• RQ1 How can process mining techniques balance formal precision with intentional vagueness to reflect uncertainty in real-world event data?
• RQ2 To what extent can hybrid Petri nets preserve formal semantics for strong causalities while representing ambiguous or complex dependencies informally?
• RQ3 Can a hybrid discovery approach scale to industrial-sized logs (e.g., millions of events) while remaining interactive and interpretable?
• RQ4 How does the proposed method compare to existing formal discovery algorithms (e.g., Inductive Miner) in terms of model quality and interpretability when silent transitions or hidden behavior are present?
• RQ5 In what ways can hybrid models improve trust and usability in process mining tools without sacrificing formal reasoning capabilities?

Key findings

• The hybrid Petri net approach successfully identifies meaningful causal relations—such as the A→B, B→C, D→E, and E→F sequences—whereas the Inductive Miner failed to express any clear causality due to hidden transitions.
• In the BPI 2016 log, only one of 18 places discovered by the Inductive Miner represented a clear causal relation, while the Hybrid Miner correctly identified the same meaningful place plus four others.
• The Hybrid Miner produced models with higher interpretability and better alignment with actual process behavior, especially in cases with complex or infrequent paths.
• The approach avoids the overconfidence of formal models by explicitly marking uncertain relations as 'unsure' arcs, preventing false precision.
• The method scales well to large logs and supports integration with existing performance and compliance analysis techniques.
• Precision metrics fail in the presence of silent transitions, highlighting the need for models that distinguish between certain and uncertain behavior—exactly what the hybrid approach provides.