Improving Process Model Precision by Loop Unrolling

Despite the advent of scalable process mining techniques that can handle both noisy and incomplete real-life event logs, there is a lack of scalable algorithms capable of handling a common cause of model underfitting: when the same activity in the log in fact behaves differently depending on the number of occurrences in a particular trace. This paper proposes a simple scalable technique to identify these cases and successfully mine better process models from event logs. The technique has been implemented and evaluated on well-known benchmarks in the literature.