Impact of Inter-Case Features on Structure-Aware Next Activity Prediction

Predictive process monitoring (PPM) techniques analyze event logs from enterprise information systems to predict the outcomes of ongoing processes. Traditional PPM approaches model process executions using intra-case features, which are based on information specific to the execution itself. However, inter-case features, which capture the broader context of the environment in which a process is executed, should also be considered. In this paper, we define new inter-case features capturing system workload in terms of concurrent cases and resources utilization. We test their impact using a Graph Neural Network architecture, which accounts for parallel activity executions. Results show that integrating inter-case features substantially improves classification performance across various real-world event logs.