Conformance Checking and Performance Improvement in Scheduled Processes: A Queueing-Network Perspective (Extended Abstract)

Arik Senderovich, Matthias Weidlich, Liron Yedidsion, Avigdor Gal, Avishai Mandelbaum, Sarah Kadish, Craig A. BunnellEMISA 2016 · OpenReview

Abstract

In complex service environments, such as oncology clinics or transportation hubs, there is often a significant "reality gap" between pre-defined schedules and actual execution. This paper introduces a framework that bridges this gap by modeling both schedules and real-world event logs as Fork/Join queueing networks. This approach enables a two-dimensional conformance checking process—conceptual and operational—to diagnose why processes deviate from plans. Furthermore, the authors propose adaptive scheduling policies that can reduce lateness and wait times by over 20% compared to traditional industry standards.

Introduction

Traditional process mining focuses heavily on the control flow—the sequence of activities. However, in resource-driven, scheduled processes, the sequence is only part of the story. The timing and resource availability defined in a schedule often conflict with the reality recorded in event logs. This research addresses two critical questions: how to mathematically assess if actual execution matches a schedule's assumptions, and how to adapt scheduling policies to improve punctuality and efficiency once gaps are identified.

The Queueing-Network Framework

The authors move away from simple trace-to-model comparisons, instead utilizing Queueing Networks as a common language to describe both planned and actual processes. Specifically, they employ Fork/Join networks to capture parallel activities and complex resource dependencies.

Overview of the proposed framework for conformance checking and performance improvement.

As shown in the framework overview, the process begins by discovering queueing network models from both the "ideal" schedule and the "messy" event logs. This allows for a two-dimensional conformance assessment:

  1. Conceptual Conformance: This compares the underlying "DNA" of the two networks—their structure, routing, and server dynamics—using statistical hypothesis testing. It determines if the fundamental assumptions about how the process works are accurate.
  2. Operational Conformance: This measures the predictive power of the schedule, asking whether the delays predicted during planning match the delays observed in reality.

Performance Improvement and Policy Adaptation

Most scheduled systems default to an Earliest-Due Date (EDD) policy. While EDD is theoretically optimal if all tasks are available simultaneously, it often fails in dynamic systems where clients arrive at staggered times. The authors argue that service processes are multi-stage interactions where improving one part of the chain without considering "Fork/Join" dependencies can lead to bottlenecks elsewhere.

To address this, the framework includes algorithms to adapt scheduling policies locally at each node in the network. By moving toward more adaptive approaches, the system can better maintain punctuality in the face of real-world variability.

Empirical Validation

The research was validated using two primary data sources:

  • Real-World Data: Real-Time Location System (RTLS) data from a large outpatient oncology clinic at the Dana-Farber Cancer Institute was used to validate the conformance checking techniques. The method successfully identified operational deviations and pinpointed their root causes.
  • Simulation Data: To evaluate the performance improvement side, simulations compared the proposed adaptive policies against standard industry practices. The results showed a reduction in both tardiness (non-punctuality) and flow time (total time in system) by more than 20%.

Conclusion and Future Work

By comparing queueing network models rather than individual traces, this framework identifies systemic root causes of delays. While the results are promising, several challenges remain for future exploration. These include the organizational hurdles of implementing algorithmic scheduling suggestions in clinical environments and the need to generalize Fork/Join models for more unpredictable sectors, such as emergency services or purely digital workflows.

Summary updated

BibTeX
@inproceedings{senderovich2016conformance,
  author = {Arik Senderovich and Matthias Weidlich and Liron Yedidsion and Avigdor Gal and Avishai Mandelbaum and Sarah Kadish and Craig A. Bunnell},
  title = {{Conformance Checking and Performance Improvement in Scheduled Processes: A Queueing-Network Perspective (Extended Abstract)}},
  booktitle = {EMISA},
  year = {2016},
  url = {https://openreview.net/forum?id=tfI32l9HPF},
}
Figure