Analysis of Business Processes with Automatic Detection of KPI Thresholds and Process Discovery Based on Trace Variants

Main Article Content

Taro Takei
Hiroki Horita


data mining; process mining; process discovery; graph edit distance.


One effective way to analyze an event log of a complex business process is to filter the event log by a KPI threshold and then use an analysis algorithm. Event logs filtered by KPI thresholds are simpler and easier to understand than the original business process. KPI thresholds have conventionally had to be set through a trial-and-error process, but existing research has proposed an automatic KPI threshold detection method that reduces the time and effort required to search for a threshold value. In the existing method, an event log is first divided by an arbitrary number k, and a business process model is generated for the divided event log using a process discovery algorithm. By repeatedly aggregating similar business process models, it is possible to analyze how the business process models differ according to KPI values. However, the existing method requires trial-and-error to determine the value of k because the detection threshold and business process model vary depending on the value of k, which is time-consuming. Therefore, this paper proposes a method to automatically detect KPI thresholds by dividing event logs based on trace variants. Experimental results show that the business process models detected by the proposed method are simpler and the threshold detection time is significantly reduced.