Designing and implementing a framework for event-based predictive modelling of business processes

Abstract

Applying predictive modelling techniques to event data collected during business process execution is receiving increasing attention in the literature. In this paper, we present a framework supporting real-time prediction for business processes. After fitting a probabilistic model to historical event data, the framework can predict how running process instances will behave in the near future, based on the behaviour seen so far. The probabilistic modelling approach is carefully designed to deliver comprehensible results that can be visualized. Thus, domain experts can judge the predictive models by comparing the visualizations to their experience. Model analysis techniques can be applied if visualizations are too complex to be understood entirely. We evaluate the framework's predictive modelling component on real-world data and demonstrate how the visualization and analysis techniques can be applied.