Relying on heterogeneous data sources to detect business process change in process models

Abstract

Due to changing customer needs, regulations, protocols, and technologies, an organi zation’s business processes must regularly change and improve. The Business Process Management (BPM) discipline guides organizations to perform these changes through the BPM life-cycle, in which business processes are modeled, analyzed, redesigned, and implemented. However, sometimes these changes bypass the BPM life-cycle, happening directly at the implementations’ operational level. Consequently, the respective process models need to be updated. Business process event logs can be analyzed to identify which models need updates, but not all implementations generate event logs. One possible approach to help detect business process changes is monitoring external sys tems, participants, documents, and other items used or produced by a business process. These items are observable entities, which are components required for a business pro cess execution. Monitoring change in these entities turns them into heterogeneous data sources, named as such because their data cannot easily be merged with event logs. We show that these entities can be used to create a framework for assisting in updating out dated process models, though it demands a method for identifying these entities. It also requires the mapping between entities and process models, allowing process analysts to quickly identify outdated models when the linked entities have suffered changes. In this thesis, we assess the feasibility of creating this framework. We evaluated and compared different frameworks of organizational change, business process analysis, and redesign with an investigation of the changes required to update 25 real process models. This comparison guided us to define a taxonomy of observable entities related to business process change, which we applied to manually classify 1329 process elements originating from 88 process models. The classification frequency of the process models was 57% on average. The classification was also used to train automated classifiers using machine learning. The best automated classifiers achieved F1-scores of up to 95.4%. Our method of semi-automated manual classification of process elements with process analysts is the primary method for identifying observable entities as required by our sug gested framework. In addition, we defined a set of recommendations to help build the mapping between entities and process models and ensure it stays consistent, as well as instructions on how to use the framework to identify outdated process models.