Inter-case Properties and Process Variant Considerations in Time Prediction: A Conceptual Framework

Predictive monitoring in business processes has gained attention in recent years. It uses a predictive model, learned from event logs, to predict the variables of interest for a running process instance (case). An example of such a variable considered here is the remaining time to complete the running case. Prediction usually relies on the properties of individual cases. Recently, the effects of the case’s environment, particularly cases that are executed in parallel to it, have been incorporated into prediction as inter-case properties. Furthermore, it has been recognized that, when different variants of the process exist, variant information should be considered by the predictive model. However, different prediction approaches use inter-case properties and variant information differently, and there is still no clear and agreed-upon manner in which these are considered for prediction. This paper proposes a conceptual framework that suggests categories of inter-case properties related to cases within a time window. Moreover, the framework considers the possible variant-awareness of these properties and suggests how variant information should be addressed in a predictive model. Reported experimentation supports our proposals.