Learning Business Rules for Adaptive Process Models

This work presents a new approach to handling knowledge-intensive business processes in an adaptive, flexible and accurate way. We propose to support processes by executing a process skeleton, consisting of the most important recurring activities of the process, through a workflow engine. This skeleton should be kept simple. The corresponding workflow is complemented by two features: firstly, a task management tool through which workflow tasks are delivered and that give human executors flexibility and freedom to adapt tasks by adding subtasks and resources as required by the context. And secondly, a component that learns business rules from the log files of this task management and that will predict subtasks and resources on the basis of knowledge from previous executions. We present supervised and unsupervised approaches for rule learning and evaluate both on a real business process with 61 instances. Results are promising, showing that meaningful rules can be learned even from this comparatively small data set.