Prescriptive Analytics for a Queuing Model without Abandonment
36 Pages Posted: 2 Feb 2020
Date Written: January 9, 2020
This paper proposes and studies novel data-driven approaches to solve a specific class of capacity sizing problems with uncertain, time varying arrival rates and patient customers that do not abandon the queue while waiting for a service. Our approaches build on a fluid approximation of the underlying queuing system and use machine learning techniques that leverage extensive auxiliary data for prescribing interrelated capacities of multiple shifts. In particular, we employ the prescriptive analytics approaches weighted sample average approximation (weighted SAA), kernelized empirical risk minimization (kernelized ERM) and an ex-post optimization approach (ExPost approach) to solve the capacity sizing problem. We demonstrate the applicability of all three approaches to the real-world staff capacity sizing problem of an aviation maintenance service provider and evaluate their performance relative to traditional approaches. The results show that prescriptive analytics approaches can lead to significant performance improvements, and that at least some are robust across variations of exogenous parameters. We observe that the prescriptive analytics approaches benefit from both their capability of exploiting predictive features, and their ability to model the time structure of arrivals with high accuracy.
Keywords: Prescriptive Analytics, Machine Learning, Data-driven OM, Capacity Management
Suggested Citation: Suggested Citation