A Dashboard-based Predictive Process Monitoring Engine

Process monitoring forms an integral part of business process management. It involves activities in which process execution data are collected and analyzed to measure the process performance with respect to the performance objectives. Traditionally, process monitoring has been performed at runtime, providing a real-time overview of the process performance and identifying performance issues as they arise. Recently, the rapid adoption of workflow management systems with logging capabilities has spawned the active development of data-driven, predictive process monitoring that exploits the historical process execution data to predict the future course of ongoing instances of a business process. Thus, potentially deviant process behavior can be anticipated and proactively addressed. To this end, various approaches have been proposed to tackle typical predictive monitoring problems, such as whether an ongoing process instance will fulfill its performance objectives, or when will an instance be completed. However, so far these approaches have largely remained in the academic domain and have not been widely applied in industry settings, mostly due to the lack of software support. In this thesis, we have designed and implemented a prototype of a predictive process monitoring engine. The developed solution, named Nirdizati, is a configurable full-stack web framework that enables the prediction of several performance indicators and is easily extensible with new predictive models for other indicators. In addition, it allows handling event streams that originate from multiple business processes. The results of the predictions, as well as the real-time summary statistics about the process execution, are presented in a dashboard that offers multiple alternative visualization options. The dashboard updates periodically based on the arriving stream of events. The solution has been successfully validated with respect to the established functional and non-functional requirements using event streams corresponding to two real-life business processes.