Discovering Hierarchical Models of Business Processes from Event Logs

Improving business processes in companies is an important and challenging task. To achieve the goal, their models should be constructed from automatically generated logs. Considering the increasing complexity of informational systems, logs are becoming more and more complicated and traditional process discovery algorithms have problems dealing with them. The discovered models are spaghetti-like, difficult to understand and may mislead analysts. The main reasons for overgeneralizing models is the fact that they are flat (without any hierarchy). Therefore a process mining topic – a mining of the hierarchical model - is becoming relevant. The main aim of the work is to propose a supervised method for the discovery of hierarchical process models from low-level system events log. The proposed method will be based on low-level log transformation. The method consist of two parts. First, we propose the algorithm for generating the high-level log from the low-level log using the algorithm for discovering high-level acyclic process models and the algorithm to inferring the repetitive behavior from event logs. The obtained log than may be used for synthesizing a high-level model by using some traditional process discovery algorithm, like inductive miner. The second part is an algorithm for finding models with low-level events that represent subprocesses and associated with high-level events in the received model. The result hierarchical model will be a target high-level model and the set of low-level models associated to every high-level event. We will proof that the proposed method guarantees fitness. Key Words: process mining, process model discovery, hierarchical models, low-level log transformations.