A Predictive Approach Enabling Process Execution Recommendations

In enterprises, decision makers need to continuously monitor business processes to guarantee for a high product and service quality. To accomplish this task, process-related data needs to be retrieved from various information systems - periodically or in real-time -- and then be aggregated based on key performance indicators (KPIs). If target values of the defined KPIs are violated (e.g., production takes longer than a predefined threshold), the reasons of these violations need to be identified. In general, such a retrospective analysis of business process data does not always contribute to prevent respective key performance violations. To remedy this drawback, process-aware information systems (PAIS) should enable the automated identification of processes, which are not well performing, and support users in executing these processes through recommendations. For example, it should be indicated, which problems might occur in future when taking the current course of the process instance as well as previous process instances into account. This chapter presents a methodology as well as an architecture for the support of predictive process analyses. In this context, algorithms from machine learning are applied to compare running process instances with historic process data and to identify diverging processes. In particular, the predictive approach will enable enterprises to quickly react to upcoming problems and inefficiencies.