Prescriptive Analytics for a Multi-Shift Staffing Problem

Abstract

Motivated by the work with a leading maintenance service provider in the aviation
industry, this paper examines novel data-driven approaches to solving a certain
type of capacity-sizing problem—the multi-shift staffing problem—with uncertain,
time varying arrival rates and patient "customers" that do not abandon the queue
while waiting for a service, but who must be served by a pre-defined time. Drawing
on established methods in both capacity management and prescriptive analytics, we
propose to use fluid and stationary approximations to apply tailored prescriptive
analytics approaches to determine staffing levels for multiple interrelated shifts. The
prescriptive analytics approaches rely on machine learning techniques that incorporate
a detailed representation of the non-stationary structure of arrivals and leverage
extensive auxiliary data that may be predictive of demand. In particular, we adapt
established prescriptive analytics approaches—weighted sample average approximation
and kernelized empirical risk minimization—and propose a new optimization
prediction approach to solving the multi-shift staffing problem. Using a case study
that is based on extensive data from our project partner, the maintenance service
provider, we demonstrate the applicability of these approaches, highlight their benefits
over traditional "estimate then optimize" approaches, and shed light on their
structural properties and performance drivers. In the context of our real-world application,
we derive a clear recommendation for the choice of method with which to
solve the multi-shift staffing problem.

Keywords: Prescriptive Analytics, Machine Learning, Data-driven OM, Queuing System