Balancing fitness and precision in process model repair: framework formalization, assessment, and benefits for simulation

Process models are essential tools for Process Science, guiding stakeholders on process automation. They are also employed to engage stakeholders into discussion to analyze how the process currently is, with the goal of discussing compliance and performance issues. They are also used as backbone to create process simulation models, with the goal of analyzing the “as-is” processes and of proposing alternative process models that guarantee improved performances. In all these settings, desirable models need to only allow all the legitimate behavior (high precision), while disallowing forbidden behavior (high fitness). Often, models fail to achieve these properties and need to be repaired. This paper proposes a model-repair framework that assembles process-simulation techniques, Machine-learning models, and Explainable AI, aiming to pinpoint the parts of the process models that enable illegitimate behavior and/or disallow valid one. These indications are then used to automatically repair a model and achieve better balances fitness and precision. Our framework has been implemented and evaluated on several models for five processes. The experimental results show that we outperform the state of the art in terms of balancing fitness and precision. The practical relevance of this technique is further showcased by illustrating the benefits of our repairing process models as backbone to create more accurate simulation models, capable to simulate more realistic runs. The simulation of more realistic process runs ensures that any process simulation-driven process improvements lead to advancements that exist in the real process world and do not remain the simulated world.