Bayesian analysis of queues with impatient customers: Applications to call centers

Queuing models have been extensively used in the literature for obtaining performance measures and developing staffing policies. However, most of this work has been from a pure probabilistic point of view and has not addressed issues of statistical inference. In this article, we consider Bayesian queuing models with impatient customers with particular emphasis on call center operations and discuss further extensions. We develop the details of Bayesian inference for queues with abandonment such as the M/M/s + M model (Erlang‐A). In doing so, we discuss the estimation of operating characteristics and its implications on staffing. We illustrate the implementation of the Bayesian models using actual arrival, service, and abandonment data from call centers. © 2012 Wiley Periodicals, Inc. Naval Research Logistics, 2012     

Stochastic call center staffing with uncertain arrival,service and abandonment rates: A Bayesian perspective

In this article, we introduce staffing strategies for the Erlang‐A queuing system in call center operations with uncertain arrival, service, and abandonment rates. In doing so, we model the system rates using gamma distributions that create randomness in operating characteristics used in the optimization formulation. We divide the day into discrete time intervals where a simulation based stochastic programming method is used to determine staffing levels. More specifically, we develop a model to select the optimal number of agents required for a given time interval by minimizing an expected cost function, which consists of agent and abandonment (opportunity) costs, while considering the service quality requirements such as the delay probability. The objective function as well as the constraints in our formulation are random variables. The novelty of our approach is to introduce a solution method for the staffing of an operation where all three system rates (arrival, service, and abandonment) are random variables. We illustrate the use of the proposed model using both real and simulated call center data. In addition, we provide solution comparisons across different formulations, consider a dynamic extension, and discuss sensitivity implications of changing constraint upper bounds as well as prior hyper‐parameters. © 2016 Wiley Periodicals, Inc. Naval Research Logistics 63: 460–478, 2016     

Service design at diagnostic service centers

We study a service design problem in diagnostic service centers, call centers that provide medical advice to patients over the phone about what the appropriate course of action is, based on the caller's symptoms. Due to the tension between increased diagnostic accuracy and the increase in waiting times more in‐depth service requires, managers face a difficult decision in determining the optimal service depth to guide the diagnostic process. The specific problem we consider models the situation when the capacity (staffing level) at the center is fixed, and when the callers have both congestion‐ and noncongestion‐related costs relating to their call. We develop a queueing model incorporating these features and find that the optimal service depth can take one of two different structures, depending on factors such as the nurses' skill level and the maximum potential demand. Sensitivity analyses of the two optimal structures show that they are quite different. In some situations, it may (or may not) be optimal for the manager to try to expand the demand at the center, and increasing skill level may (or may not) increase congestion. © 2012 Wiley Periodicals, Inc. Naval Research Logistics, 2012     

Sequential learning versus no learning in Bayesian regression models

We consider two regression models: linear and logistic. The dependent variable is observed periodically and in each period a Bayesian formulation is used to generate updated forecasts of the dependent variable as new data is observed. One would expect that including new data in the Bayesian updates results in improved forecasts over not including the new data. Our findings indicate that this is not always true. We show there exists a subset of the independent variable space that we call the “region of no learning.” If the independent variable values for a given period in the future are in this region, then the forecast does not change with any new data. Moreover, if the independent variable values are in a neighborhood of the region of no learning, then there may be little benefit to wait for the new data and update the forecast. We propose a statistical approach to characterize this neighborhood which we call the “region of little learning.” Our results provide insights into the trade‐offs that exist in situations when the decision maker has an incentive to make an early decision based on an early forecast versus waiting to make a later decision based on an updated forecast. © 2014 Wiley Periodicals, Inc. Naval Research Logistics 61: 532–548, 2014     

An integer programming model for the weekly tour scheduling problem

We study a workforce planning and scheduling problem in which weekly tours of agents must be designed. Our motivation for this study comes from a call center application where agents serve customers in response to incoming phone calls. Similar to many other applications in the services industry, the demand for service in call centers varies significantly within a day and among days of the week. In our model, a weekly tour of an agent consists of five daily shifts and two days off, where daily shifts within a tour may be different from each other. The starting times of any two consecutive shifts, however, may not differ by more than a specified bound. Furthermore, a tour must also satisfy constraints regarding the days off, for example, it may be required that one of the days off is on a weekend day. The objective is to determine a collection of weekly tours that satisfy the demand for agents' services, while minimizing the total labor cost of the workforce. We describe an integer programming model where a weekly tour is obtained by combining seven daily shift scheduling models and days‐off constraints in a network flow framework. The model is flexible and can accommodate different daily models with varying levels of detail. It readily handles different days‐off rules and constraints regarding start time differentials in consecutive days. Computational results are also presented. © 2001 John Wiley & Sons, Inc. Naval Research Logistics 48: 607–624, 2001.     

Machine scheduling with an availability constraint and job delivery coordination

In this paper we study the scheduling problem that considers both production and job delivery at the same time with machine availability considerations. Only one vehicle is available to deliver jobs in a fixed transportation time to a distribution center. The vehicle can load at most K jobs as a delivery batch in one shipment due to the vehicle capacity constraint. The objective is to minimize the arrival time of the last delivery batch to the distribution center. Since machines may not always be available over the production period in real life due to preventive maintenance, we incorporate machine availability into the models. Three scenarios of the problem are studied. For the problem in which the jobs are processed on a single machine and the jobs interrupted by the unavailable machine interval are resumable, we provide a polynomial algorithm to solve the problem optimally. For the problem in which the jobs are processed on a single machine and the interrupted jobs are nonresumable, we first show that the problem is NP‐hard. We then propose a heuristic with a worst‐case error bound of 1/2 and show that the bound is tight. For the problem in which the jobs are processed on either one of two parallel machines, where only one machine has an unavailable interval and the interrupted jobs are resumable, we propose a heuristic with a worst‐case error bound of 2/3. © 2006 Wiley Periodicals, Inc. Naval Research Logistics, 2007     

Stochastic framework for partially degradation systems with continuous component degradation‐rate‐interactions

Many conventional models that characterize the reliability of multicomponent systems are developed on the premise that for a given system, the failures of its components are independent. Although this facilitates mathematical tractability, it may constitute a significant departure from what really takes place. In many real‐world applications, system components exhibit various degrees of interdependencies, which present significant challenges in predicting degradation performance and the remaining lifetimes of the individual components as well as the system at large. We focus on modeling the performance of interdependent components of networked systems that exhibit interactive degradation processes. Specifically, we focus on how the performance level of one component affects the degradation rates of other dependent components. This is achieved by using stochastic models to characterize how degradation‐based sensor signals associated with the components evolve over time. We consider “Continuous‐Type” component interactions that occur continuously over time. This type of degradation interaction exists in many applications, in which interdependencies occur on a continuum. We use a system of stochastic differential equations to capture such “Continuous‐Type” interaction. In addition, we utilize a Bayesian approach to update the proposed model using real‐time sensor signals observed in the field and provide more accurate estimation of component residual lifetimes. © 2014 Wiley Periodicals, Inc. Naval Research Logistics 61: 286–303, 2014     

Bayesian inference for a Lanchester type combat model

We undertake inference for a stochastic form of the Lanchester combat model. In particular, given battle data, we assess the type of battle that occurred and whether or not it makes any difference to the number of casualties if an army is attacking or defending. Our approach is Bayesian and we use modern computational techniques to fit the model. We illustrate our method using data from the Ardennes campaign. We compare our results with previous analyses of these data by Bracken and Fricker. Our conclusions are somewhat different to those of Bracken. Where he suggests that a linear law is appropriate, we show that the logarithmic or linear‐logarithmic laws fit better. We note however that the basic Lanchester modeling assumptions do not hold for the Ardennes data. Using Fricker's modified data, we show that although his “super‐logarithmic” law fits best, the linear, linear‐logarithmic, and logarithmic laws cannot be ruled out. We suggest that Bayesian methods can be used to make inference for battles in progress. We point out a number of advantages: Prior information from experts or previous battles can be incorporated; predictions of future casualties are easily made; more complex models can be analysed using stochastic simulation techniques. © 2000 John Wiley & Sons, Inc. Naval Research Logistics 47: 541–558, 2000     

Equilibrium customer strategies and social–profit maximization in the single‐server constant retrial queue

We consider the single‐server constant retrial queue with a Poisson arrival process and exponential service and retrial times. This system has not waiting space, so the customers that find the server busy are forced to abandon the system, but they can leave their contact details. Hence, after a service completion, the server seeks for a customer among those that have unsuccessfully applied for service but left their contact details, at a constant retrial rate. We assume that the arriving customers that find the server busy decide whether to leave their contact details or to balk based on a natural reward‐cost structure, which incorporates their desire for service as well as their unwillingness to wait. We examine the customers' behavior, and we identify the Nash equilibrium joining strategies. We also study the corresponding social and profit maximization problems. We consider separately the observable case where the customers get informed about the number of customers waiting for service and the unobservable case where they do not receive this information. Several extensions of the model are also discussed. © 2011 Wiley Periodicals, Inc. Naval Research Logistics, 2011     

Production‐inventory systems with imperfect advance demand information and updating

We consider a supplier with finite production capacity and stochastic production times. Customers provide advance demand information (ADI) to the supplier by announcing orders ahead of their due dates. However, this information is not perfect, and customers may request an order be fulfilled prior to or later than the expected due date. Customers update the status of their orders, but the time between consecutive updates is random. We formulate the production‐control problem as a continuous‐time Markov decision process and prove there is an optimal state‐dependent base‐stock policy, where the base‐stock levels depend upon the numbers of orders at various stages of update. In addition, we derive results on the sensitivity of the state‐dependent base‐stock levels to the number of orders in each stage of update. In a numerical study, we examine the benefit of ADI, and find that it is most valuable to the supplier when the time between updates is moderate. We also consider the impact of holding and backorder costs, numbers of updates, and the fraction of customers that provide ADI. In addition, we find that while ADI is always beneficial to the supplier, this may not be the case for the customers who provide the ADI. © 2011 Wiley Periodicals, Inc. Naval Research Logistics, 2011     

A robust policy for serial agile production systems

One of the major problems in modeling production systems is how to treat the job arrival process. Restrictive assumptions such as Markovian arrivals do not represent real world systems, especially if the arrival process is generated by job departures from upstream workstations. Under these circumstances, cost‐effective policies that are robust with respect to the nature of the arrival process become of interest. In this paper, we focus on minimizing the expected total holding and setup costs in a two‐stage produce‐to‐order production system operated by a cross‐trained worker. We will show that if setup times are insignificant in comparison with processing times, then near‐optimal policies can be generated with very robust performances with respect to the arrival process. We also present conditions under which these near‐optimal policies can be obtained by using only the arrival and service rates. © 2004 Wiley Periodicals, Inc. Naval Research Logistics, 2005.     

On‐line algorithms for minimizing makespan on batch processing machines

We consider problem of scheduling jobs on‐line on batch processing machines with dynamic job arrivals to minimize makespan. A batch machine can handle up to B jobs simultaneously. The jobs that are processed together from a batch, and all jobs in a batch start and complete at the same time. The processing time of a batch is given by the longest processing time of any job in the batch. Each job becomes available at its arrival time, which is unknown in advance, and its processing time becomes known upon its arrival. In the first part of this paper, we address the single batch processing machine scheduling problem. First we deal with two variants: the unbounded model where B is sufficiently large and the bounded model where jobs have two distinct arrival times. For both variants, we provide on‐line algorithms with worst‐case ratio (the inverse of the Golden ratio) and prove that these results are the best possible. Furthermore, we generalize our algorithms to the general case and show a worst‐case ratio of 2. We then consider the unbounded case for parallel batch processing machine scheduling. Lower bound are given, and two on‐line algorithms are presented. © 2001 John Wiley & Sons, Inc. Naval Research Logistics 48: 241–258, 2001     

A note on optimal pricing for finite capacity queueing systems with multiple customer classes

This article investigates optimal static prices for a finite capacity queueing system serving customers from different classes. We first show that the original multi‐class formulation in which the price for each class is a decision variable can be reformulated as a single dimensional problem with the total load as the decision variable. Using this alternative formulation, we prove an upper bound for the optimal arrival rates for a fairly large class of queueing systems and provide sufficient conditions that ensure the existence of a unique optimal arrival rate vector. We show that these conditions hold for M/M/1/m and M/G/s/s systems and prove structural results on the relationships between the optimal arrival rates and system capacity. © 2008 Wiley Periodicals, Inc. Naval Research Logistics, 2008     

Signature based analysis of m‐Consecutive‐k‐out‐of‐n: F systems with exchangeable components

In this article, we study reliability properties of m‐consecutive‐k‐out‐of‐n: F systems with exchangeable components. We deduce exact formulae and recurrence relations for the signature of the system. Closed form expressions for the survival function and the lifetime distribution as a mixture of the distribution of order statistics are established as well. These representations facilitate the computation of several reliability characteristics of the system for a given exchangeable joint distribution or survival function. Finally, we provide signature‐based stochastic ordering results for the system's lifetime and investigate the IFR preservation property under the formulation of m‐consecutive‐k‐out‐of‐n: F systems. © 2011 Wiley Periodicals, Inc. Naval Research Logistics, 2011     

Optimal policies under the shortage probability criterion for an inventory model with unknown dependent demands

The principal innovation in this paper is the consideration of a new objective function for inventory models which we call the shortage probability criterion. Under this criterion we seek to minimize the total expected discounted cost of ordering subject to the probability that the stock level at the end of the period being less than some fixed quantity not exceed some prescribed number. For three different models we show that the minimum order policy is optimal. This result is then applied to a particular inventory model in which the demand distribution is not completely known. A Bayesian procedure is discussed for obtaining optimal policies.     

Natural gas storage valuation and optimization: A real options application

In this article, we present an algorithm for the valuation and optimal operation of natural gas storage facilities. Real options theory is used to derive nonlinear partial‐integro‐differential equations (PIDEs), the solution of which give both valuation and optimal operating strategies for these facilities. The equations are designed to incorporate a wide class of spot price models that can exhibit the same time‐dependent, mean‐reverting dynamics, and price spikes as those observed in most energy markets. Particular attention is paid to the operational characteristics of real storage units. These characteristics include working gas capacities, variable deliverability and injection rates, and cycling limitations. We illustrate the model with a numerical example of a salt cavern storage facility that clearly shows how a gas storage facility is like a financial straddle with both put and call properties. Depending on the amount of gas in storage the relative influence of the put and call components vary. © 2009 Wiley Periodicals, Inc. Naval Research Logistics 2009     

Optimization and mediated bartering models for ground delay programs

The Federal Aviation Administration (FAA) and the airline community within the United States have adopted a new paradigm for air traffic flow management, called Collaborative Decision Making (CDM). A principal goal of CDM is shared decision‐making responsibility between the FAA and airlines, so as to increase airline control over decisions that involve economic tradeoffs. So far, CDM has primarily led to enhancements in the implementation of Ground Delay Programs, by changing procedures for allocating slots to airlines and exchanging slots between airlines. In this paper, we discuss how these procedures may be formalized through appropriately defined optimization models. In addition, we describe how inter‐airline slot exchanges may be viewed as a bartering process, in which each “round” of bartering requires the solution of an optimization problem. We compare the resulting optimization problem with the current procedure for exchanging slots and discuss possibilities for increased decision‐making capabilities by the airlines. © 2005 Wiley Periodicals, Inc. Naval Research Logistics, 2006     

Aggregating courier deliveries

We consider the problem of efficiently scheduling deliveries by an uncapacitated courier from a central location under online arrivals. We consider both adversary‐controlled and Poisson arrival processes. In the adversarial setting we provide a randomized (3βΔ/2δ ? 1) ‐competitive algorithm, where β is the approximation ratio of the traveling salesman problem, δ is the minimum distance between the central location and any customer, and Δ is the length of the optimal traveling salesman tour overall customer locations and the central location. We provide instances showing that this analysis is tight. We also prove a 1 + 0.271Δ/δ lower‐bound on the competitive ratio of any algorithm in this setting. In the Poisson setting, we relax our assumption of deterministic travel times by assuming that travel times are distributed with a mean equal to the excursion length. We prove that optimal policies in this setting follow a threshold structure and describe this structure. For the half‐line metric space we bound the performance of the randomized algorithm in the Poisson setting, and show through numerical experiments that the performance of the algorithm is often much better than this bound.     

Dynamic service rate control for a single‐server queue with Markov‐modulated arrivals

We consider the problem of service rate control of a single‐server queueing system with a finite‐state Markov‐modulated Poisson arrival process. We show that the optimal service rate is nondecreasing in the number of customers in the system; higher congestion levels warrant higher service rates. On the contrary, however, we show that the optimal service rate is not necessarily monotone in the current arrival rate. If the modulating process satisfies a stochastic monotonicity property, the monotonicity is recovered. We examine several heuristics and show where heuristics are reasonable substitutes for the optimal control. None of the heuristics perform well in all the regimes and the fluctuation rate of the modulating process plays an important role in deciding the right heuristic. Second, we discuss when the Markov‐modulated Poisson process with service rate control can act as a heuristic itself to approximate the control of a system with a periodic nonhomogeneous Poisson arrival process. Not only is the current model of interest in the control of Internet or mobile networks with bursty traffic, but it is also useful in providing a tractable alternative for the control of service centers with nonstationary arrival rates. © 2013 Wiley Periodicals, Inc. Naval Research Logistics 60: 661–677, 2013     

A mathematical formulation and efficient heuristics for the dynamic container relocation problem

The container relocation problem (CRP) is concerned with emptying a single yard‐bay which contains J containers each following a given pickup order so as to minimize the total number of relocations made during their retrieval process. The CRP can be modeled as a binary integer programming (IP) problem and is known to be NP‐hard. In this work, we focus on an extension of the CRP to the case where containers are both received and retrieved from a single yard‐bay, and call it the dynamic container relocation problem. The arrival (departure) sequences of containers to (from) the yard‐bay is assumed to be known a priori. A binary IP formulation is presented for the problem. Then, we propose three types of heuristic methods: index based heuristics, heuristics using the binary IP formulation, and a beam search heuristic. Computational experiments are performed on an extensive set of randomly generated test instances. Our results show that beam search heuristic is very efficient and performs better than the other heuristic methods.Copyright © 2014 Wiley Periodicals, Inc. Naval Research Logistics 61: 101–118, 2014