An Effective Process Mining Approach against Diverse Logs Based on Case Classification

Since real-life processes tend to be much flexible because of the ever changing circumstances, there is a lot of diversity in logs leading to complex models which may contain various kinds of complex control-flow structures. However, every mining algorithm has its pros and cons, so there is not a general algorithm which is capable to handle diverse logs. In this paper, we propose a general process mining approach, which first deals with the diversity issue by classifying the cases into sets of categories (sub logs). Next, multiple process miners take these sub logs as input to produce sets of process models. Then, a genetic algorithm (GA) based optimizer taking these process models as parts of initial population aggregates appropriate process fragments into the entire process model with the balance of four quality dimensions. Experiments on synthetic and real-life logs from a telecommunication giant demonstrate the effectiveness of our approach.