Inductive Miner with Clustering of Fall-Through Sublogs

Extracting interpretable business process models from event logs of poorly structured (spaghetti) processes is challenging. The Inductive Miner algorithm is a popular process mining algorithm. However, it tends to produce fall-through nodes when applied to spaghetti processes, resulting in a model with many stacked activities, which makes it harder to interpret the underlying process. In this work, we propose an adaptation to the Inductive Miner algorithm, called IM_Cluster, that clusters the traces of the event log associated with fall-through nodes, aiming to produce fewer occurrences of stacking of activities. We also devised a new simplicity metric that can capture the occurrence of stacking of activities, and the IM_Cluster relies on this metric. An empirical evaluation is conducted to compare the proposed algorithm with the original Inductive Miner and the ActiTraC algorithm. The results show that the proposed algorithm produces fewer occurrences of stacking of activities than the compared algorithms.