Machine Learning for Non-Markovian Queueing Analysis
Dr. Eliran Sherzer, Industrial Engineering, Ariel University.
Abstract:
Classical queueing theory has long provided fundamental tools for analyzing service systems, yet its practical reach is often constrained by strong analytical assumptions. When arrival processes, service times, or network structures deviate from Markovian settings, steady-state analysis quickly becomes analytically intractable and computationally intensive, relying almost exclusively on large-scale simulation.
This work introduces a supervised machine-learning paradigm that fundamentally changes how steady-state behavior in queueing systems can be analyzed. Drawing on two studies, Supervised ML for Solving the GI/GI/1 Queue and Computing the Steady-State Probabilities of Tandem Queueing Systems: A Machine Learning Approach, neural networks are trained on simulation-generated data to learn direct mappings from system characteristics to stationary performance measures.
A central insight of this approach is the use of moment-based representations of general interarrival and service-time distributions, coupled with a systematic moment analysis that reveals how a small number of low-order moments can capture much of the system’s long-run behavior. Once trained, the models deliver near-instantaneous predictions of steady-state distributions, effectively replacing repeated simulation runs.
Taken together, these results suggest a new direction for queueing analysis, one in which machine learning complements classical theory to overcome long-standing analytical barriers. By enabling fast, accurate, and scalable steady-state inference in complex queueing models, this framework has the potential to significantly expand the scope and impact of queueing theory in modern service and production systems.
Bio:
Eliran Sherzer earned his BSc, MSc, and PhD in Industrial Engineering from Ben-Gurion University of the Negev, and continued his research training as a postdoctoral fellow at Tel Aviv University and the University of Toronto. He is currently a researcher in the Department of Industrial Engineering at Ariel University. His research combines queueing theory and machine learning to develop scalable methods for analyzing and optimizing complex stochastic systems.
