Dolores Romero Morales

Integrated Lot Sizing in Serial Supply Chains with Production Capacities

We consider a model for a serial supply chain in which production, inventory, and transportation decisions are integrated in the presence of production capacities and concave cost functions. The model we study generalizes the uncapacitated serial single-item multilevel economic lot-sizing model by adding stationary production capacities at the manufacturer level. We present algorithms with a running time that is polynomial in the planning horizon when all cost functions are concave. In addition, we consider different transportation and inventory holding cost structures that yield improved running times: inventory holding cost functions that are linear and transportation cost functions that are either linear, or are concave with a fixed-charge structure. In the latter case, we make the additional common and reasonable assumption that the variable transportation and inventory costs are such that holding inventories at higher levels in the supply chain is more attractive from a variable cost perspective. While the running times of the algorithms are exponential in the number of levels in the supply chain in the general concave cost case, the running times are remarkably insensitive to the number of levels for the other two cost structures.

Supervised classification and mathematical optimization

Data mining techniques often ask for the resolution of optimization problems. Supervised classification, and, in particular, support vector machines, can be seen as a paradigmatic instance. In this paper, some links between mathematical optimization methods and supervised classification are emphasized. It is shown that many different areas of mathematical optimization play a central role in off-the-shelf supervised classification methods. Moreover, mathematical optimization turns out to be extremely useful to address important issues in classification, such as identifying relevant variables, improving the interpretability of classifiers or dealing with vagueness/noise in the data.

A class of greedy algorithms for the generalized assignment problem

The Generalized Assignment Problem (GAP) is the problem of finding the minimal cost assignment of jobs to machines such that each job is assigned to exactly one machine, subject to capacity restrictions on the machines. We propose a class of greedy algorithms for the GAP. A family of weight functions is defined to measure a pseudo-cost of assigning a job to a machine. This weight function in turn is used to measure the desirability of assigning each job to each of the machines. The greedy algorithm then schedules jobs according to a decreasing order of desirability. A relationship with the partial solution given by the LP-relaxation of the GAP is found, and we derive conditions under which the algorithm is asymptotically optimal in a probabilistic sense.

A Branch-and-Price Algorithm for the Multiperiod Single-Sourcing Problem

In this paper, we propose a multiperiod single-sourcing problem (MPSSP), which takes both transportation and inventory into consideration, suitable for evaluating the performance of a logistics distribution network in a dynamic environment. We reformulate the MPSSP as a Generalized Assignment Problem (GAP) with a convex objective function. We then extend a branch-and-price algorithm that was developed for the GAP to this problem. The pricing problem is a so-called Penalized Knapsack Problem (PKP), which is a knapsack problem where the objective function includes an additional convex term penalizing the total use of capacity of the knapsack. The optimal solution of the relaxation of the integrality constraints in the PKP shows a similar structure to the optimal solution of the knapsack problem, that allows for an efficient solution procedure for the pricing problem. We perform an extensive numerical study of the branch-and-price algorithm.

A Heuristic Approach to the Multi-Period Single-Sourcing Problem with Production and Inventory Capacities and Perishability Constraints

The multi-period single-sourcing problem that we address in this paper can be used as a tool for evaluating logistics network designs in a dynamic environment. We consider the assignment of retailers to facilities, taking into account the timing, location, and size of production and inventories, in the presence of various types of constraints. We formulate the problem as a nonlinear assignment problem, and develop efficient algorithms for solving the capacitated lot-sizing subproblems that form the objective function of this formulation. We propose a greedy heuristic, and prove that this heuristic is asymptotically optimal in a probabilistic sense when retailer demands share a common seasonality pattern. In addition, we develop an efficient implementation of the very-large-scale-neighborhood-search method that can be used to improve the greedy solution. We perform extensive tests on a set of randomly generated problem instances, and conclude that our approach produces very high quality solutions in limited time.

Forecasting cancellation rates for services booking revenue management using data mining

Revenue management (RM) enhances the revenues of a company by means of demand-management decisions. An RM system must take into account the possibility that a booking may be canceled, or that a booked customer may fail to show up at the time of service (no-show). We review the Passenger Name Record data mining based cancellation rate forecasting models proposed in the literature, which mainly address the no-show case. Using a real-world dataset, we illustrate how the set of relevant variables to describe cancellation behavior is very different in different stages of the booking horizon, which not only confirms the dynamic aspect of this problem, but will also help revenue managers better understand the drivers of cancellation. Finally, we examine the performance of the state-of-the-art data mining methods when applied to Passenger Name Record based cancellation rate forecasting.

Binarized Support Vector Machines

The widely used support vector machine (SVM) method has shown to yield very good results in supervised classification problems. Other methods such as classification trees have become more popular among practitioners than SVM thanks to their interpretability, which is an important issue in data mining. In this work, we propose an SVM-based method that automatically detects the most important predictor variables and the role they play in the classifier. In particular, the proposed method is able to detect those values and intervals that are critical for the classification. The method involves the optimization of a linear programming problem in the spirit of the Lasso method with a large number of decision variables. The numerical experience reported shows that a rather direct use of the standard column generation strategy leads to a classification method that, in terms of classification ability, is competitive against the standard linear SVM and classification trees. Moreover, the proposed method is robust; i.e., it is stable in the presence of outliers and invariant to change of scale or measurement units of the predictor variables. When the complexity of the classifier is an important issue, a wrapper feature selection method is applied, yielding simpler but still competitive classifiers.

A nested heuristic for parameter tuning in Support Vector Machines

The default approach for tuning the parameters of a Support Vector Machine (SVM) is a grid search in the parameter space. Different metaheuristics have been recently proposed as a more efficient alternative, but they have only shown to be useful in models with a low number of parameters. Complex models, involving many parameters, can be seen as extensions of simpler and easy-to-tune models, yielding a nested sequence of models of increasing complexity. In this paper we propose an algorithm which successfully exploits this nested property, with two main advantages versus the state of the art. First, our framework is general enough to allow one to address, with the very same method, several popular SVM parameter models encountered in the literature. Second, as algorithmic requirements we only need either an SVM library or any routine for the minimization of convex quadratic functions under linear constraints. In the computational study, we address Multiple Kernel Learning tuning problems for which grid search clearly would be infeasible, while our classification accuracy is comparable to that of ad hoc model-dependent benchmark tuning methods.

Asymptotic Analysis of a Greedy Heuristic for the Multi-Period Single-Sourcing Problem: The Acyclic Case

The multi-period single-sourcing problem that we address in this paper can be used as a tactical tool for evaluating logistics network designs in a dynamic environment. In particular, our objective is to find an assignment of customers to facilities, as well as the location, timing and size of production and inventory levels, that minimizes total assignment, production, and inventory costs. We propose a greedy heuristic, and prove that this greedy heuristic is asymptotically optimal in a probabilistic sense for the subclass of problems where the assignment of customers to facilities is allowed to vary over time. In addition, we prove a similar result for the subclass of problems where each customer needs to be assigned to the same facility over the planning horizon, and where the demand for each customer exhibits the same seasonality pattern. We illustrate the behavior of the greedy heuristic, as well as some improvements where the greedy heuristic is used as the starting point of a local interchange procedure, on a set of randomly generated test problems. These results suggest that the greedy heuristic may be asymptotically optimal even for the cases that we were unable to analyze theoretically.

Generating Experimental Data for the Generalized Assignment Problem

The Generalized Assignment Problem GAP is the problem of finding the minimal cost assignment of jobs to machines such that each job is assigned to exactly one machine, subject to capacity restrictions on the machines. We propose a new stochastic model for the GAP. A tight condition on this stochastic model under which the GAP is feasible with probability one when the number of jobs goes to infinity is derived. This new stochastic model enables us to analyze the adequacy of most of the random generators given for the GAP in the literature. We demonstrate that the random generators commonly used to test solution procedures for the GAP tend to create easier problem instances when the number of machines m increases. We describe a greedy heuristic for the GAP, and use it to illustrate the results from the paper.