Explainable Artificial Intelligence for Improved Modeling of Processes

Machine Learning can handle complex business data better than traditional methods: While traditional process modeling techniques create easy-to-understand diagrams, they struggle with real-world data that comes from many different sources. The paper shows that modern ML models, particularly Transformer architectures, can work with this complex, high-dimensional data more effectively.

AI explanations reveal what actually matters in business processes: The study demonstrates that by using explainability techniques (attention mechanisms and feature relevance analysis), organizations can identify which specific factors are most important for predicting critical business outcomes—providing "new insights into the underlying processes" that wouldn't be visible through traditional analysis alone.

Predictive accuracy and interpretability can work together: The research validates that ML models don't have to be "black boxes"—the paper shows these models can both make accurate predictions about process outcomes AND explain which features drove those predictions, addressing a key concern in responsible AI implementation.

Testing on multiple datasets strengthens the findings: The authors validated their approach across five different benchmark datasets, suggesting the results are robust and potentially applicable across different types of business processes rather than being limited to one specific context.