Event Data and Process Model Forecasting

Process mining studies ways to use event data generated by information systems to understand and improve the business processes of organizations. One of the core problems in process mining is process discovery. A process discovery algorithm takes event data as input and constructs a process model that describes the processes the system that generated the data can execute. The discovered model, hence, aims to represent both historical processes with traces in the data and the yet unseen processes of the system (total generalization). In this paper, we introduce process forecasting as an alternative approach to process discovery. First, given historical event data, the corresponding future event data is forecasted for a requested period in the future (event data forecasting). Then, a process model is constructed from the forecasted data to describe the processes the system is anticipated to execute during the target future period (process model forecasting). The benefits of this alternative approach are at least twofold. Firstly, it divides the problem into two fundamentally different sub-problems that can be studied and mastered separately. Secondly, a forecasted model that describes the processes of the system from a given period rather than in general (tailored generalization) can help organizations plan future operations and process improvement initiatives.