A genetic algorithm for discovering process trees

Abstract

Existing process discovery approaches have problems dealing with competing quality dimensions (fitness, simplicity, generalization, and precision) and may produce anomalous process models (e.g., deadlocking models). In this paper we propose a new genetic process mining algorithm that discovers process models from event logs. The tree representation ensures the soundness of the model. Moreover, as experiments show, it is possible to balance the different quality dimensions. Our genetic process mining algorithm is the first algorithm where the search process can be guided by preferences of the user while ensuring correctness.

Data-Driven Process Discovery - Revealing Conditional Infrequent Behavior from Event Logs

“…Most of the more recent and more sophisticated process discovery methods support noise filtering [3]. Existing noise-filtering methods are based on frequencies [7,8,9,10], machine-learning techniques [11,12], genetic algorithms [13], or probabilistic models [14,15]. All of those methods focus on the controlflow perspective (i.e., the event labels) when filtering noise.…”

Improving Business Process Models Using Observed Behavior

“…The algorithm stops after 1, 000 generations or sooner if a candidate with perfect overall fitness is found before. In [7] we empirically showed that 1, 000 generations are typically enough to find the optimal solution, especially for processes with few activities. All other settings were selected according to the optimal values presented in [7].…”

Discovering Block-Structured Process Models from Incomplete Event Logs

“…They provide a natural, structured, well-defined way of describing processes that are often easily translatable to Petri nets. The process tree formalisms used in [8,17,18] guarantee soundness as well. Process tree discovery techniques have also been proposed before.…”

“…Process trees, or block structures in general, have been used in process discovery, both inside the scope of Petri nets [8,2,20], as outside [24,25] the scope of Petri nets. They provide a natural, structured, well-defined way of describing processes that are often easily translatable to Petri nets.…”

“…For instance, the approach used by [26] constructs a process tree from a log by enumerating all traces, after which the process tree is simplified. The Evolutionary Tree Miner (ETM) [8] uses a genetic approach to discover a process tree, i.e., a random population is mutated until a certain stop criterion is met, but as it is steered by log-based metrics, fitness, precision, generalisation and simplicity, and by its random nature, it is unable to guarantee rediscoverability. A natural strategy when using block structures is to apply a divide-and-conquer strategy, which has been applied to process discovery in for instance [9,36,17,18].…”