Enhancing remaining time prediction in business processes by considering system-level and resource-level inter-case features

Accurately predicting the remaining time of running cases is crucial for effective scheduling in business processes. To achieve accurate predictions, it is essential to consider the conditions and behavior of organizational resources, which significantly impact process completion times. Interactions between different cases can change resource conditions, so inter-case features created by these interactions should be considered in remaining time prediction. These inter-case features can be considered at both the system and resource levels. While past studies have largely focused on system-level features, they have often neglected the impact of resource-level features. This research investigates the effect of inter-case features on various prediction models by developing a conceptual framework that extracts open cases and resource multitasking as indicators of system and resource workloads, along with resource experience features, from event logs. Using four regression algorithms, four bucketing methods, and five encoding techniques, the study applies predictive process monitoring models to eight real-world datasets. The findings indicate that incorporating inter-case features generally improves prediction accuracy, particularly in processes with high resource and system workloads. However, the optimal model configuration for the highest accuracy varies across datasets. Although inter-case features enhance prediction accuracy, they also increase both offline and online execution times. This study underscores the importance of considering both system and resource workloads for accurate remaining time prediction and provides a framework to guide the integration of inter-case features.