Using Sequential Pattern Mining and Social Network Analysis to Identify Similarities, Differences and Evolving Behaviour in Event Logs

This paper reports on findings resulting from our research conducted for the 2015 Business Process Intelligence Challenge (BPIC), an annual competition in which participants are tasked with conducting process mining-related analysis on a real-life dataset. This year’s data was provided by 5 Dutch municipalities, and contained activity pertaining to their building permit application process. Questions regarding the underlying process posed to the participants centered around identifying differences in control flow among municipalities, the responsible factors for longer processing times, and differences in the various roles of employees involved, to name a few. Our approach to addressing these questions involved the application of methods from the field of sequential pattern mining, an area of research that identifies frequently occurring sequences of events in potentially large databases. In particular, sequence classification is used to identify statistically significant differences in control flow among municipalities. Also, value-based sequential pattern mining is used to identify patterns in control flow that are linked to 1) high/low throughput times, in order to identify similarities and differences among the five municipalities, and 2) earlier/later process instances, in order to examine how municipalities’ underlying process may have changed over time, and how these changes may be similar among municipalities. We also employ traditional methods from the field of process mining to shed light on the the social network-related aspects of the data, such as how the roles of employees differ among municipalities in terms of task similarity.