Umeå University's logo

In complex information systems, assessing whether a sequence of activities - referred to as a case or process instance - complies with behavioral rules is crucial. This assessment, known as conformance checking, involves specifying declarative constraints on process behavior, which are converted into database queries and executed using specialized query engines. However, interpreting the results of numerous conformance-checking queries on large datasets poses significant challenges. To enhance explainability, this thesis introduces formal notions of concise conformance violation explanations that summarize these results and evaluate their feasibility using real-world data. Specifically, the thesis defines minimal necessary constraint violations explanations and counterfactual explanations as part of a broader framework for explainability. Additionally, the framework introduces contribution functions to quantify the impact of constraints and behavioral variants on fitness and conformance loss.

The framework is integrated and tested with an industry-scale query language, providing interpretable insights into process conformance violations. The empirical results indicate minimal explanations are computationally efficient, while counterfactuals are more computationally heavy. Contribution functions provide a structured and fair assessment of constraint and variant impacts, making the framework applicable to large-scale event logs, although with certain scalability concerns.