Business process simulation is a well-known approach to estimate the impact of changes to a process with respect to time and cost measures – a practice known as what-if process analysis. The usefulness of such estimations hinges on the accuracy of the underlying simulation model. Data-Driven Simulation (DDS) methods leverage process mining techniques to learn business process simulation models from event logs. Empirical studies have shown that, while DDS models adequately capture the observed sequences of activities and their frequencies, they fail to accurately capture the temporal dynamics of real-life processes. In contrast, generative Deep Learning (DL) models are better able to capture such temporal dynamics. The drawback of DL models is that users cannot alter them for what-if analysis due to their black-box nature. This paper presents a hybrid approach to learn process simulation models from event logs wherein a (stochastic) process model is extracted via DDS techniques, and then combined with a DL model to generate timestamped event sequences. The proposed approach allows us to simulate different types of changes, including the addition of new activity types to a process. This latter capability is achieved by encoding the activities by means of embeddings, rather than representing them as one-hot-encoded categories. An experimental evaluation shows that the resulting hybrid simulation models match the temporal accuracy of pure DL models, while partially retaining the what-if analysis capability of DDS approaches. The evaluation also sheds light into the relative performance of multiple embedding approaches to represent the activities.

Abstract Business process simulation is a well-known approach to estimate the impact of changes to a process with respect to time and cost measures – a practice known as what-if process analysis. The usefulness of such estimations hinges on the accuracy of the underlying simulation model. Data-Driven Simulation (DDS) methods leverage process mining techniques to learn process simulation models from event logs. Empirical studies have shown that, while DDS models adequately capture the observed sequences of activities and their frequencies, they fail to accurately capture the temporal dynamics of real-life processes. In contrast, generative Deep Learning (DL) models are better able to capture such temporal dynamics. The drawback of DL models is that users cannot alter them for what-if analysis due to their black-box nature. This paper presents a hybrid approach to learn process simulation models from event logs wherein a (stochastic) process model is extracted via DDS techniques, and then combined with a DL model to generate timestamped event sequences. An experimental evaluation shows that the resulting hybrid simulation models match the temporal accuracy of pure DL models, while partially retaining the what-if analysis capability of DDS approaches.

Business process simulation is a versatile technique for quantitative analysis of business processes. A well-known limitation of process simulation is that the accuracy of the simulation results is limited by the faithfulness of the process model and simulation parameters given as input to the simulator. To tackle this limitation, various authors have proposed to discover simulation models from process execution logs, so that the resulting simulation models more closely match reality. However, existing techniques in this field make certain assumptions about resource behavior that do not typically hold in practice, including: (i) that each resource performs one task at a time; and (ii) that resources are continuously available (24/7). In reality, resources may engage in multitasking behavior and they work only during certain periods of the day or the week. This article proposes an approach to discover process simulation models from execution logs in the presence of multitasking and availability constraints. To account for multitasking, we adjust the processing times of tasks in such a way that executing the multitasked tasks sequentially with the adjusted times is equivalent to executing them concurrently with the original times. Meanwhile, to account for availability constraints, we use an algorithm for discovering calendar expressions from collections of time-points to infer resource timetables from an execution log. We then adjust the parameters of this algorithm to maximize the similarity between the simulated log and the original one. We evaluate the approach using real-life and synthetic datasets. The results show that the approach improves the accuracy of simulation models discovered from execution logs both in the presence of multitasking and availability constraints