Current approaches to process mining lack robustness to noisy data. In this paper, a hybrid approach is proposed for process mining. The objective of the algorithm is to achieve robustness, as well as to provide global search. It uses well-defined Petri net notations for the genetic algorithm. Different metrics for checking the accuracy are defined. The authors show that the algorithm can mine sequence, choice, parallelism, loops, and invisible tasks, along with being robust to noise, which is quite comprehensive compared with other current approaches.

It is unclear, however, whether the two algorithms selected for comparison are sufficient. A comparison with other algorithms out of the ProM framework could have been made to further improve the conclusions on preciseness and completeness. The extremely high cost in terms of time constitutes a big deficiency. Since the post-processing can be easily overtuned, it can result in decreased accuracy; if situations that involve bad running time, along with overtuning, happen, the performance could really be worse.