ABSTRACT. We present a method for non-smooth convex minimization based on subgradient directions affected by random errors and on String-Averaging (SA) techniques. Incremental subgradient algorithms can be viewed as decentralized network optimization methods applied to minimize a sum of local functions, which are known only to a particular agent of a distributed network. The addition of the SA methods in this context can be interpreted as an algorithmic version of a network in which subnets process independent sequences being coordinated by a central node, which calculates and distributes the averages. We consider stochastic errors in the form of additive noise in the subgradient and we assume that the second moments of the subgradient noise are uniformly bounded over the agents, in such a manner that the errors do not necessarily decrease over time. The main result of this paper consists in a theoretical convergence analysis of the proposed algorithm when we use a diminishing step-size rule.

ABSTRACT. En este trabajo se presenta un algoritmo basado en direcciones factibles del tipo direcciones de Zoutendijk para resolver problemas de programación lineal. Si bien existen algoritmos eficientes en la práctica para este problema, como el algoritmo simplex por ejemplo, resulta interesante explorar un algoritmo como este, basado en direcciones, en que el número de iteraciones está asociado al tamaño del problema de manera lineal. En los problemas resueltos se encontró que el número de iteraciones del algoritmo expresado a través de la secuencia de puntos generados fue siempre menor o igual que el número de variables del problema. Esto podría dar lugar a rutinas eficientes para problemas de gran tamaño.

ABSTRACT. A significant amount of applied problems depend on the minimization of nonsmooth functions. Consequently, many optimization methods have been developed in order to deal with these kind of functions. Recently, a method known as Gradient Sampling (GS) has gained attention by its good results (specially on difficult problems). One may view the GS method as a generalization of the steepest descent method. In fact, when we apply the GS algorithm to a smooth function, its search direction asymptotically converges to the steepest descent direction. Despite many studies have been made from the original GS ideas, for the best of our knowledge, none of them proved a local convergence result. Since the GS method locally approximates the objective function by sampling some points around the current iterate, a local result is not trivial and will be intrinsically linked to a good set of sampled points. Therefore, this study has the intent to present a local convergence result for the GS method.

ABSTRACT. Let $(x^k)$ be a sequence generated by a typical first-order method, intended to minimize a closed convex function $f:H\to\R\cup\{+\infty\}$ on a Hilbert space $H$ (think of $\R^N$). If $f$ has minimizers, then $(x^k)$ converges (weakly, if $H$ is infinite-dimensional) to one of them, with a worst-case theoretical convergence rate of $f(x^k)-\min(f)=O(1/k)$. In the 1980's Nesterov conceived a simple but powerful idea to modify the computation of the iterates with the same computational cost, but a convergence rate of $O(1/k^2)$. For decades, researchers tried unsuccessfully to prove the convergence of the resulting iterates. Despite this drawback, Nesterov's scheme is current common practice in imaging and statistics, and converges faster than the unaccelerated counterpart. It turns out that a slight variant Nesterov’s method yields convergent sequences with a convergence rate of $f(x^k)-\min(f)=o(1/k)$. The convergence of the original method of Nesterov remains open.

ABSTRACT. This research addresses single and multi commodity periodic review based inventory-location problems, for a three level supply chain with stochastic demands.Decisions involve warehouse location, retailer-warehouse assignments and inventory control parameters for each warehouse (reorder point and order size). Decisions are optimized ensuring inventory control service level and stochastic capacity constraints. In this research, the use of Lagrangian Relaxation and sub-gradient method are proposed, considering the relaxation of different set of constraints, including customer assignment, and warehouses demand and variance constraints. In addition, we develop a simple local search heuristic to achieve feasible solutions at each algorithm step. Computational results for several instances provide small duality gaps, near optimal solutions, and low computational times, where significant impacts of the inventory control policy into system configuration and costs can be observed.

ABSTRACT. En productos con poca vida útil, como las frutas y hortalizas, o bien, en aquellos que requieren refrigeración, como las carnes, la frescura es un factor determinante para garantizar su calidad. A pesar de la existencia de modelos heurísticos, no se reportan modelos que puedan ser utilizados para resolver este tipo de problema utilizando metaheurísticas. El presente estudio aborda el problema de ruteo de vehículos refrigerantes y de tipo general para la entrega de un producto perecedero, conociendo la demanda, ubicación de clientes, tiempos de servicio, tiempos de viaje entre clientes, capacidad y número de vehículos disponibles de ambos tipos. Se busca reducir la pérdida de frescura del producto entregado, considerando el tiempo transcurrido desde el depósito hasta el cliente y las aperturas de puerta de los vehículos en ruta. Se utilizan 3 metaheurísticas para resolver este problema, logrando establecer la más adecuada para esta variante de VRP

ABSTRACT. We address the distribution logistics of a small-sized dairy company in Northern Brazil, which produces and delivers products to retailers spread in a relatively large area. Due to the distances involved, each route may comprise several days before the trucks return to the company´s warehouse. In addition, periodic long stops in charging stations are required to recharge the battery of the equipment that keeps the temperature in the trucks bodies within acceptable levels. In this study, a mixed integer linear model was developed to describe the problem as a multi-period vehicle routing problem with time windows and periodic stops. In addition, a relax-and-fix heuristic was applied to solve the model with real and generated instances up to 40 costumers.

ABSTRACT. Several small and medium companies have facilities far away from urban area. Due to most employees don’t have easy access to their jobs, companies are in the need to hire a transport service for its staff. Poor urban planning and traffic congestion, especially in emerging countries, led to excessive delays in worker's transport time and decreases their quality of life. This waiting time is even greater if the company transportation system is inefficient. Our approach seeks to optimize the companies’ transport service by determining the frequency and size of vehicles. A simulation model is proposed to minimize user’s waiting time. This approach is validated using real-data taken from Colombia. Results put in evidence the quantitative benefits that can be achieved when the model is implemented, represented in both service level and transportation costs. Additionally, different alternatives for decision makers are presented, in order to economically evaluate different service levels.

ABSTRACT. La logística humanitaria se reconoce como un campo de estudio que aborda diversos aspectos logísticos y operacionales que suceden en el contexto de la gestión del riesgo de desastres. Una de las actividades que generan mayor atención en éste ámbito, corresponde a la asistencia hacia los posibles afectados desde un punto de vista logístico. La presente investigación aplicada aborda la generación de rutas óptimas de evacuación ante un tsunami, en conjunto con la posterior localización de albergues y centros de abastecimiento en la ciudad de Talcahuano, Chile, mediante un algoritmo de programación matemática basado en p-medianas capacitado y un sistema de información geográfico (SIG), cuyo objetivo corresponde a la minimización de los tiempos totales de viaje. Finalmente, se efectuó el análisis de los resultados obtenidos en virtud de instancias de aplicación con datos y escenarios reales en esta comuna.

ABSTRACT. Para resolver um problema integrado de dimensionamento de lotes e rotamento de veículos típico de empresas moveleiras de pequeno porte é proposto um modelo de programação inteira mista. O modelo inclui características próprias do setor, como produção e estocagem de itens, rotas de distribuição estendendo-se por vários períodos, múltiplas janelas de tempo e prazos de entrega. Testes computacionais mostraram que o modelo representa de forma apropriada a situação real. Além disto, testes realizados com base em um conjunto de exemplares gerados aleatoriamente, mostram que o CPLEX 12.6 encontra soluções ótimas ou factíveis para exemplares de até n=10 clientes em tempos computacionais razoáveis.

ABSTRACT. In 2015, Leisten and Rajendran introduced the variability-related objective function CTDVB2 which aims at smoothness of a flowshop’s output by minimizing the difference of completion times of two consecutively scheduled jobs. In this work, CTDVB2 and modifications of CTDVB2 are investigated in serial assembly systems. The influence of variability in one system on the overall performance of the combined system is evaluated via computational experiments. First, a single flow shop with transport unit is evaluated from a scheduling perspective. Then, two flowshops in series with and without intra-shop logistics are analysed from a queueing perspective. Looking at the overall performance from scheduling and queueing perspectives helps to close the gap between static-deterministic scheduling and dynamic-stochastic queueing theory.

ABSTRACT. The situation here addressed is modeled as a bilevel programming problem with multiple objectives in the upper level and a single objective in the lower level. In this problem, a company (hereafter the leader) distribute a product over a selected subset of clients; while a manufacturer (hereafter the follower) will fabricate the products demanded by the selected clients. The leader has two objectives: the maximization of the profit gained by the distribution process and the minimization of CO2 emissions. The latter is important due to the regulations imposed by the government. It is clear that exists a compromise between both objectives, since the maximization of profit will attempt to include as much clients for being served as possible. Then, largest routes will be needed causing more CO2 emissions. A Nash-Genetic algorithm for solving the aforementioned problem is proposed. The algorithm is designed for the obtaining bilevel feasible solutions; that is, the follower’s problem is optimally solved by a commercial optimizer for each leader’s decision. The Nash approach for the algorithm is related with having two separate populations, one for each leader’s objective. Then, the solutions will evolve in each population and a co-evolutionary scheme is periodically included. In other words, a swap between the best solutions in each generation is conducted. By doing this, the algorithm intends to find the efficient solutions. The evolution performed in each population is done through Biased Random Keys (BRK). Under this scheme a solution is represented by a random key of length 3N + 1 -where N are the number of clients- that contains random numbers between 0 and 1. The designed decoder and the conceptualization of the algorithm are detailed. In addition for the feasibility of the random keys a local search to convert the infeasible random keys to feasible for the initial population is designed. A battery of instances for validating and measuring the proposed methodology were generated. Computational experimentation of the proposed algorithm, analysis of the obtained results and a properly visualization of the findings are exposed.

ABSTRACT. Desde a sua criação, a indústria cinematográfica tem produzido uma grande variedade de filmes e a cada lançamento várias informações são disponibilizadas à população em geral. Sem muito tratamento, essas informações não chegam de forma acessível e personalizada. Com o desenvolvimento dos sistemas de recomendação, a maneira com que as pessoas compram pela internet mudou consideravelmente. Em meio a uma gama de opções disponíveis, o usuário passou a ter um auxílio, capaz de guiá-lo até um produto que possa interessá-lo. Assim, o objetivo deste trabalho é calibrar um modelo capaz de prover um sistema colaborativo que vai sugerir um determinado filme para um usuário do serviço baseado na avaliação efetuada por usuários de interesses similares.

ABSTRACT. For better results, the quality of the specialists consulted to generate the weights of the AHP must be relevant. Independently of the quality from the specialists, it is possible to identify some bias in each of the responses, sometimes creating a group that obey some sort of characteristics. The main goal of this work is to identify or, in the perfect scenario pre-identify, those bias in order to modify the model, through the translation of the weights matrix, using the eigenvalues to do so, changing the final order generated by the AHP method. This approach is useful for the decision maker because it can show the final order by class of specialist and its interests. For example, in the case study, it was identified three bias (groups) in the responses, named as industry, academia and regulation. The industry wants a model that results in a higher profit, the academia aims the better model that fit the data and the regulation is concerned about the consequences at the environment.

ABSTRACT. The parameter estimation in logistic regression model is known to be dependent on the data configuration. While the logistic model work well for many situations, may not work when the groups of observations are completely separated. Separation is a common problem in the logistic regression. Mathematical Programming approaches have been used for detecting separated data in logistic regression, but most of these researches have focused on the two group problem. In this paper we propose a linear programming formulation to detect separation in polytomous logistic regression. The proposed approach classifies data as completely separated, quasi-separated or overlapped, and can be used as part of the parameter estimation. Comparison with other methods, using different data sets taken from the literature, shows that our formulation may suggest an efficient alternative to mathematical programming approaches for the multiple-group problem.

ABSTRACT. Pantaneiro horses were submitted to a performance test. Performance was measured by a data envelopment analysis – DEA model – with four outputs and one unit input. The output measures were the distance achieved in the performance test, hematocrit as a weighted average over the test duration, cardiac and respiratory frequencies as weighted averages over the test duration and the level of lactic acid at the test termination. Contextual variables of interest are age, horse weight, room temperature and corporal temperature, and presence or absence of a condition known as equine infectious anemia (EIA). Only the EIA effect and room temperature affect performance statistically significantly, as indicated by a nonparametric replication analysis.The techniques used in the analysis can be extended, with little changes, to other applications, whether or not in the agricultural sector.

ABSTRACT. Let D = (V,A) be a digraph with a set of vertices V and a set of arcs A, with c_ij  representing the cost of an arc (i,j) in A. The problem of finding the shortest elementary path in the presence of negative cycles (SEPPNC) is NP-hard and consists in determining, if it exists, the path of minimum cost between two distinguished vertices s in V and t in V. We propose three solution approaches for the SEPPNC, which are a compact primal-dual formulation, a branch-and-bound algorithm, and a cutting-plane method. An extensive computational study performed on both benchmark and randomly generated instances shows that our approaches outperform state-of-the-art solution techniques for this problem, while providing optimal solutions for benchmark instances in very small execution times.

ABSTRACT. Las redes de telecomunicaciones supervivientes son aquellas que siguen funcionando a pesar de fallas. El objetivo es diseñar las redes de forma tal que se pueda garantizar la proteccion del trafico frente a ciertos tipos de fallas al menor costo posible. La reciente tecnologia de p-ciclos se convirtio rapidamente en una tecnica prometedora debido a que brinda los beneficios combinados de la velocidad de recuperacion de los anillos y la eficiencia de la malla. En las redes basadas en p-ciclos, cada ciclo protege ante la falla de un link que forma parte del ciclo o de uno que tiene sus dos extremos en el ciclo. En este trabajo propusimos varios modelos de programacion entera y mixta para el problema basico de definir una topologia basada en p-cyclos optimizando los costos (SCA, Spare Capacity Allocation) y describimos un algoritmo Branch and Price basado en los mismos. Tambien se propuso un metodo basado en programacion por restricciones.

ABSTRACT. p-Cycles are structures intended to provide survivability to optical networks. Each p-Cycle provides one protection channel to each span it crosses, and two protection channels to each span that is not in the cycle but its ending nodes are. The Spare Capacity Allocation problem requires protecting all working demands against any span failure with p-cycles at minimum cost. We propose here a greedy heuristic that builds a solution for this problem iteratively. At each step, a Genetic Algorithm builds a cycle trying to maximize the Actual Efficiency, which was defined in the literature for developing another greedy heuristic for this problem. To achieve this, we propose to decode cycles from 0-1 genes representing fundamental cycles of a basis of cycle space. We give two alternatives for fitness evaluation to treat disjoint cycles or closed walks with repeated nodes. Several computational experiments were performed and promising results were obtained.

ABSTRACT. We study several inventory policies in the design of a distribution network for fast-moving items able to provide differentiated service levels in terms of product availability for two demand classes (high and low priority). In particular, we consider the distribution network design problem when the Separate Stock, Single Class Allocation, Global Round-Up, Local Round-Up and Critical Level inventory policies are used. We show how to formulate these problems as conic quadratic mixed-integer problems. For the instances solved in the computational experiments, the best policy was never the one that assigns to each distribution center a unique demand class, i.e., the single class allocation is the worst policy performance in terms of total cost.

ABSTRACT. O objetivo deste artigo é encontrar o método mais adequado de cálculo para a Provisão de Sinistros Ocorridos mas Não Avisados (IBNR - Incurred But Not Reported), para um Operadora de Plano de Saúde já estabelecida no mercado brasileiro, através da análise multicritério. Para este estudo, foi aplicado o método AHP (Analytic Hierarchy Process) e, as alternativas consideradas foram baseadas na bibliografia específica: Método proposto pela Agencia Nacional de Saúde, Método Chain Ladder, Método Bornhuetter-Ferguson, Método Benktander-Hovinen, Método Bootstrap e Método Log-Normal. Os critérios da análise foram: Cenário Passado Próprio, Sinistralidade do Mercado, Possibilidade de Simulação e Complexidade dos Cálculos. Os especialistas foram Atuários com experiência na área de Provisões Técnicas e Solvência. Os resultados encontrados apontaram para o método Log-Normal, seguido do método Bootstrap (métodos estocásticos).

ABSTRACT. A precificação e qualidade das aeronaves enquanto garantia de financiamentos tendem a ser um problema para o BNDES, principalmente, quando da devolução ou retomada de uma aeronave por parte do Banco, em casos de inadimplência. O modelo permite qualificar, comparar e precificar aeronaves baseado em características qualitativas retiradas quando da observação das reais condições da aeronave. É utilizado o VFB para estruturar uma problemática de decisão e o método AHP com Ratings para comparar as aeronaves entre si e com aeronaves teóricas precificadas pelo mercado. Assim tem-se uma ferramenta de monitoramento da carteira aeronáutica, propiciando um melhor acompanhamento das condições de manutenção das aeronaves, principais garantias dos financiamentos, por meio da classificação das aeronaves em faixas de acordo com o rating obtido, permitindo o acompanhamento dos ativos nas empresas aéreas através de relatórios qualitativos se antecipando eventuais eventos de inadimplemento.

ABSTRACT. Abstract This article addresses the problem of measuring closeness in weighted environments(decision-making environments), using the concept of compatibility of priority vectors and value systems. When using the concept of closeness come in mind immediately what means to be close (when close really means close). Thus when measuring closeness or proximity we should add a point of comparison (a threshold) that make possible to compare or decide if our positions, system values or priorities are really close.

ABSTRACT. De forma periódica, algunas compañías realizan la programación de las visitas del personal que realiza la reposición de sus productos de los puntos de venta que posee. Esta programación incluye el secuenciamiento de visita de los puntos de venta diariamente, así como, la jornadas laboral de cada reponedor. Tanto el secuenciamiento de visitas, como el diseño de las jornadas laborales deben considerar una serie de requerimientos operativos y propios del negocio, transformando esta programación en una tarea compleja de realizar manualmente. Este trabajo presenta un enfoque de solución basado en programación entera que permite determinar la programación óptima del personal que realiza la reposición de productos. Este enfoque fue aplicado a una empresa real, mejorando de manera sustancial la cobertura de los puntos de venta y los costos de contratación. Utilizando la misma dotación de personal fue pasar de un 68% de cobertura a más de un 88% con la solución del enfoque propuesto.

ABSTRACT. In July 2015, Ronaldo and Forlán drew balls from two pots to define the schedule of the South American World Cup Qualifiers to the 2018 football World Cup Russia. One pot contained the names of the ten national teams that participate in this qualification tournament. The other pot contained ten numbers referring to the position that each team would take in a predefined schedule template. We generated this schedule template by using integer linear programming. We included several criteria, such as symmetry and breaks within double rounds. This talk reports on several symmetric formats we considered as alternative to satisfy these criteria. We also compare our finally adopted solution against a previous schedule that was used in the four previous World Cup qualifiers. To the best of our knowledge, this was the first time that OR-based sports scheduling has been applied to an international competition.

ABSTRACT. This paper aims to integrate two variations of the one-dimensional cutting stock problem: the cutting stock problem with usable leftovers and the multiperiod cutting stock problem, and find a solution to the problem resulting from this integration. The cutting stock problem with usable leftovers consist of cutting larger objects into smaller sizes in order to meet a known demand, minimizing the waste and allowing generate retails that will be use to cut future items. In the multiperiod cutting stock problem demands occur in a finite period of time, being possible to cut in advance a stored object to produce items that have known demands in a future period of time. A mathematical model is proposed for the multiperiod cutting stock problem with usable leftover and is being solved using simplex method with column generation. Some preliminary computational results are presented with a simulation where each period is considered individually.

ABSTRACT. Este trabajo presenta una nueva metodología para estructurar las habilidades polifuncionales de un conjunto de empleados que inicialmente sólo cuentan con una función especialista. Asumiendo una demanda estocástica y una modelación continua de la fuerza laboral, la metodología descompone el problema de polifuncionalidad en tres etapas y lo resuelve de forma secuencial. La primera etapa entrega una expresión analítica simple para estimar la dotación de empleados polifuncionales requerida en cada actividad del sistema. La segunda etapa aplica una heurística constructiva a estas estimaciones para generar un conjunto factible de cadenas cerradas para el sistema. La tercera etapa usa simulación Montecarlo y un modelo de programación lineal para evaluar el desempeño de las cadenas cerradas generadas. Como resultado se obtienen estructuras de polifuncionalidad tipo cadenas cerradas que son robustas ante la variabilidad de la demanda, minimizan el costo total esperado de subdotación, y exhiben el mejor desempeño a nivel costo-efectivo. La metodología también entrega a los tomadores de decisiones políticas de polifuncionalidad que responden a tres preguntas fundamentales sobre la polifuncionalidad en sistemas desbalanceados: dónde, cuánta, y cómo agregarla.

ABSTRACT. La Innovación Empresarial, como todo ciclo de Gestión, requiere de una fase de evaluación que permita conocer el estado de cualquier Empresa y posibilite efectuar cambios para mejorar su administración y gestión. Dichas empresas por más pequeñas que sean, como por ejemplo una pequeña y mediana empresa (PYMES) busca la manera de permanecer en el tiempo, lo cual pueden lograrlo si optimizan sus recursos e incluso pueden obtener un incremento en sus utilidades. El trabajo que se plantea a consideración es la determinación de las prioridades competitivas de las PYMES en Quito-Ecuador; acorde a uno de los modelos establecidos por la administración de operaciones, el mismo que busca analizar qué aspectos de estas empresas pueden ser considerados como prioridades para competir con empresas de similares características o superiores en aspectos como producción, personal.

ABSTRACT. Among all the multiplicity of SMEs, technological ones distinguish by innovative approach to entrepreneurship and increased risk – by scope and level as well. Collaboration risk belongs to the most specific risk categories for this kind of activities. In the paper authors describe various areas where collaboration risk in TSMEs may influence business approach and performance, identify and analyse sources of collaboration risk, and propose rational typology for that – considerably wide – risk category. Report is a part of a broader research project re. risk in TSMEs in Poland.

ABSTRACT. The identification of factors that affect the creation of risk cooperation is a problem both in theoretical and practical terms. This study shows what impact risk cooperation have on the future of technological entrepreneurship. The paper developed the following research questions: Q1. What factors determine risk cooperation? Q2. How does risk cooperation influence the future of technology entrepreneurship? Based on the 275-element sample (data collected in 2014) and using structural equation models, a model was constructed of the impact on risk cooperation. The results confirm that the perception of decision risk is a very important in technological entrepreneurship.

ABSTRACT. La capacidad de absorción es la habilidad de la empresa para reconocer el valor de la información externa, asimilarla y aplicarla con fines comerciales. El resultado innovador, se puede dar a través de dos estrategias: complementariedad y sustituibilidad. Si es la primera hay capacidad de absorción, si es la segunda no. El reto es identificar la estrategia de innovación predominante de las empresas a través de la modelización de su capacidad de absorción. Este trabajo presenta los resultados de la modelización de la capacidad de absorción para las empresas españolas identificando la estrategia predominante de innovación, utilizando Redes Neuronales Artificiales.

ABSTRACT. We present an algorithm to solve multistage mixed 0-1 stochastic problems with nonlinear convex objective function and convex constraints. These problems have continuous and binary variables in each stage. The algorithm is based on the Branch-and-Fix Coordination (BFC) method. The non-anticipativity constraints are satisfied by means of the twin-node family strategy. In this work to solve each nonlinear convex subproblem generated at each node of the trees of the BFC method we propose the solution of sequences of quadratic subproblems. As constraints are convex we can approximate them by means of outer linear approximations. The algorithm has been implemented in C++ with the help of Cplex 12.1 to solve quadratic approximations. Test problems have been randomly generated by a C++ code. Numerical experiments have been performed and its efficiency has been compared with that of BONMIN.

ABSTRACT. Dado un grafo G = (V;E), decimos que un subconjunto A de V es P3-convexo si todo todos los vértices con dos ó más vecinos en A, deben pertenecer a A. La cápsula P3-convexa de un subconjunto B de V es el menor conjunto A (con respecto a la inclusión) tal que B está incluído en A y A es P3- convexo. El P3-hull number de G es el tamaño del menor conjunto B incluído en V tal que la cápsula P3-convexa de B es V. Calcular el hull number de un grafo es un problema NP-hard, y ha sido objeto de interés por parte de la comunidad de optimización combinatoria. En este trabajo comenzamos un estudio poliedral de este problema, a partir de una formulación natural del problema como un modelo de programación lineal entera. Estudiamos la dimensión del poliedro asociado y determinamos bajo que condiciones las restricciones del modelo denen facetas de este poliedro. Además, estudiamos familias de desigualdades válidas y analizamos bajo que condiciones estas desigualdades denen facetas de este poliedro.

ABSTRACT. Este trabalho aborda o Problema de Inundação em Grafos, cujo objetivo é tornar monocromático um grafo colorido em vértices com o menor número de inundações possível. Por inundação, entende-se a mudança da cor de um vértice pivô para uma cor c, o que faz com que esse vértice seja agregado a todos os vértices vizinhos que tenham a mesma cor c. A literatura do problema apresenta uma formulação matemática na forma de um Programa Inteiro Misto. Uma nova formulação matemática é proposta e experimentos computacionais mostraram que esta formulação tem desempenho muito superior à da literatura.

ABSTRACT. A b-vertex coloring of a graph G is a proper vertex coloring of G such that each color class contains a vertex that has at least one neighbor in every other color class. The b-chromatic number v(G) of a graph G is the largest integer k for which G has a b-vertex coloring with k colors. Integer linear programming (ILP) has been successfully applied to the classic vertex coloring problem. Nevertheless, only a few of the many variations of the classic vertex coloring have been studied in the literature under an ILP approach. In this work, we aim at analyzing whether ILP can be indeed a useful tool for the computation of v(G) in the b-vertex coloring problem. To this end, we extend some known ILP-formulations for the classical vertex coloring problem. We provide several families of valid inequalities, and we study conditions under which these inequalities define facets of the associated polytope. We show finally preliminary computational experiments for our results.

ABSTRACT. We propose a decomposition procedure based on a three-level column generation framework for the optimal delivering of industrial gases. The industrial problem can be modelled as a split-delivery routing problem plus some constraints associated the delivery of gases. The lowest-level problem computes several cargo patterns for each route generated on the intermediate-level routes-generator problem. Routes and cargo patterns form the so-called cargo-routes that are coordinated on the top-level master-problem to fulfil the demands of different gases by customers. We tested two cargo-generation procedures: via column generation leading to a nested procedure and via a procedure that implicitly generates all cargos as a convex combination of extreme cargo-patterns. Both strategies were evaluated by solving several instances derived from a real-world case including 3 plants producing LIN and LOX destined to satisfy the demand of up to 50 clients. All were solved in less than an hour.

ABSTRACT. In this work, we address the problem of customer and distribution districting. In this problem, the objective is to assign clients to a commercial zone satisfying certain districting conditions (e.g., compactness, delivery time, etc.). Our approach is a hybrid technique that combines a k-means clustering algorithm with a set covering problem with side constraints. We test our approach through a case study that consists of 3.600 clients of a major food distribution company in the Valparaiso region.

ABSTRACT. En este trabajo se optimiza la red de distribución de combustible desde una Terminal de Distribución a un conjunto de Estaciones de Servicio ubicadas al norte de la Ciudad de México. La empresa actualmente distribuye el combustible con una flota homogénea y desea saber si la inclusión de una flota heterogénea minimiza los costos de distribución e inventario, respetando las restricciones de almacenamiento y capacidad de oferta. Para la optimización se propone la recopilación de datos, análisis de la demanda, cálculo de costos de transporte e inventario y un modelo matemático utilizando el método heurístico Greedy Randomized Adaptive Search Procedure (GRASP) para su solución. Finalmente se analizan los resultados donde se logra un ahorro total de un 84.4% en relación a la situación actual, ya que disminuyen cantidades de combustible a ordenar y en la periodicidad de la distribución, por lo que se rechaza la hipótesis de que la flota heterogénea minimizaría los costos de distribución.

ABSTRACT. Los Bancos de Alimentos forman la columna vertebral en la lucha contra el hambre, han surgido con el fin proveer productos alimenticios a la población que se encuentra en pobreza extrema y hambruna. Sin embargo, al no contar con recursos suficientes para cubrir toda la demanda de productos, se tiene que determinar qué conjunto de beneficiarios serán servidos, así como las cantidades y tipos de productos que les serán asignados. Además, la asignación debe cumplir características especiales como: requerimientos de nutrición balanceada, inventarios de productos, presupuesto de compras, entre otras. Se muestran los resultados computacionales obtenidos con el branch-and-bound de CPLEX sobre un gran conjunto de instancias del problema y se presenta una heurística con el fin de obtener soluciones de buena calidad en un tiempo computacional razonable.

ABSTRACT. El sistema portuario de Valparaíso se encuentra conformado por dos elementos principales: ZEAL y las terminales portuarias. Actualmente se presentan ineficiencias en los procesos logísticos relacionados con los tiempos de estadía de camiones en el interior del sistema portuario. Las ineficiencias causan demoras que se ven reflejadas y repercuten en el desempeño de ZEAL. Las características del mercado internacional y con el objetivo de mejorar los servicios brindas es necesario contar con herramientas que permitan visualizar diferentes escenarios y realizar una evaluación de los mismos dando soporte a los procesos de toma de decisión tendientes a mejorar los indicadores logísticos. En el presente trabajo se presenta un modelo de simulación a partir del cual se puede reducir los tiempos de ejecución de los procesos aportando eficiencia a la cadena logística portuaria, la cual genera una ventaja competitiva para el puerto y los otros actores vinculados en la cadena logística portuaria.

ABSTRACT. El balance entre el estado de la malla vial local de Bogotá y la capacidad disponible del Gobierno de la ciudad para atenderla, motivaron un proceso de planeación de largo plazo que permita superar las dificultades del rezago de casi 30 años en su mantenimiento y conservación (hoy, un 56% se encuentra en mal estado) mediante una premisa de territorialización del accionar. Para afrontar esta necesidad se planteó un modelo de programación matemática que busca sectorizar la demanda de servicios en un número finito de distritos de conservación con un mínimo de desviación entre los tamaños obtenidos para cada uno, garantizando similar esfuerzo de gestión y simplificación de las tareas de asignación de recursos y evaluación de desempeño. Simultáneamente, se propone un modelo complementario de ajuste de la oferta en razón a la demanda de servicios para la generación de escenarios de decisión de largo plazo que den sosteniblidad al modelo financiero del esquema de intervención.

ABSTRACT. In this work, we propose a mathematical model for the railway rapid transit network design problem. As main input data we consider an underlying network, that is, a set of potential stations and their connections, the mobility patterns and a set of parameters related to costs and the network planning horizon. We assume the existence of a competing mode (private car, bus) as an alternative to the railway system. Because of the computational intractability of this model, we incorporate some efficient solution approaches as well as an analysis about the effect of considering different objectives functions. Some numerical experiments are carried out in order to illustrate the model performance.

ABSTRACT. En sistemas de transporte público con altos niveles de demanda, los servicios expresos son una alternativa atractiva tanto para usuarios como para operadores. Leiva et al. (2010) and Larraín et al. (2013) introducen un modelo de optimización de frecuencias que trabaja con un set dado de servicios expresos en un corredor. Estos modelos contemplan la optimización de un modelo no lineal cuya solución resulta altamente costosa en términos de tiempo y muchas veces poco confiable. Para abordar este problema se propone un esquema bi-nivel para resolver el problema de optimización de frecuencias y servicios expreso. Este enfoque además de ser más rápido en su solución, permite probar distintos modelos de asignación de usuarios y hace más eficiente la búsqueda de soluciones satisfaciendo restricciones de capacidad en los buses. Pruebas preliminares muestran que el algoritmo propuesto converge y se espera mostrar que el nuevo método permite resolver para corredores de mayor tamaño y complejidad.

ABSTRACT. Clustering is one of the main data mining tasks with many proven techniques and successful real-world applications. However, in changing environments, the existing systems need to be regularly updated in order to describe in the best possible way an observed phenomenon at each moment of time. This motivates the development of tools for dynamic clustering. Since changes lead to uncertainty, the respective systems require an adequate modeling of the involved kinds of uncertainty. We present the novel method for dynamic clustering called Dynamic Rough-Fuzzy Support Vector Clustering focusing in the mathematical properties of the methods used. The main idea is to take advantage of the knowledge acquired in previous cycles to speed up the updating of each stage of the algorithm while tracking the structural changes that clusters can experience over time. The computational experiments highlight the potential that these kinds of algorithms could have in dynamic environments.

ABSTRACT. Costa and Machado (2011) proposed the RMAX chart to control the covariance matrix of two quality characteristics. The monitoring statistic of the RMAX chart is the maximum of the two standardized sample ranges from the bivariate observations of the two quality characteristics. In this article, we investigate the performance of two synthetic RMAX. The first synthetic chart signals when a second point, not far from the first one, falls beyond the warning limit. The second synthetic chart additionally signals when a sample point falls beyond the control limit. The performance of the synthetic RMAX charts are compared with the performance of the pure RMAX chart.

ABSTRACT. El crecimiento de los préstamos entre particulares vía Internet mediante plataformas P2P es notable. En estos préstamos los inversores particulares asumen el riesgo de crédito y no las instituciones financieras. Es importante determinar la probabilidad de insolvencia y sus factores explicativos pero también predecir la rentabilidad de cada préstamo: los préstamos más rentables no siempre son los más solventes sino a veces aquellos que pagan intereses elevados. El trabajo va más allá de los sistemas orientados a predecir la insolvencia (credit scoring) y propone predecir la rentabilidad de la inversión (profit scoring). Se compara la capacidad predictiva de ocho técnicas estadísticas de minería de datos. El estudio empírico utiliza datos de Lending Club, la plataforma líder. La técnica con mejor capacidad predictiva es Support Vector Regression seguida de un modelo de red neuronal. El profit scoring mejora los resultados obtenidos con respecto a un credit scoring tradicional.

ABSTRACT. Due to the rapid advances in information technology at deteriorating costs, the amount of data that has been collected multiplied in the last decade. For companies it has become crucial to use this data to remain competitive in a global economy. To address this challenge, concepts like Business Analytics, Data Science, besides others, are intensively discussed currently. Clustering is a frequently used approach in data mining and one of the methods to tackle these issues. A particular challenge arises when data change over time, i.e., the data domain is not stable. For such situations dynamic clustering methods have been developed. Recently, the Dynamic Clustering Cube Framework (DCC) has been proposed to categorize dynamic clustering and disclose methodically gaps that should be considered to be filled. Putting the aspects mentioned above together, the objective of our research is to discuss and evaluate the potential of the DCC-Framework for data mining applications in business.

ABSTRACT. In this talk a hierarchized Steiner's tree problem in networks is presented. Users need to be connected to capacitated hubs, then the selected hubs will be connected among them and with some extra hubs -if necessary-. Connection between hubs can be seen as obtaining a Steiner's tree. The problem is modelled as a bi-level mathematical programming problem where two decision levels are considered. In the upper level, the allocation of users to hubs is made aiming the minimization of the total network's connection cost. On the other hand, the lower level consists in minimizing the user's latency with respect to the information flowing through the capacitated hubs. A co-evolutionary algorithm is developed for solving the problem using parallel computing in one of the populations. An application for the topological design of a LAN is studied. Numerical results show the convenience of the proposed approach when the lower level problem cannot be optimally solved in an efficient manner.

ABSTRACT. En confiabilidad de redes, se desea hallar la probabilidad de correcta operaci\'on de una red sujeta a fallas. En este art\'iculo se estudia un modelo hostil de red, donde pueden fallar tanto los enlaces como los nodos en forma independiente. Tomaremos como hip\'otesis de trabajo la independencia de fallas de los componentes no terminales, y funcionamiento perfecto de sus terminales. La red opera correctamente siempre que los terminales (nodos distinguidos de la misma) permanezcan comunicados. Este modelo generaliza el problema cl\'asico de confiabilidad de redes, y pertenece como consecuencia a la clase de problemas $\mathcal{NP}$-Dif\'iciles. Brindamos en este art\'iculo m\'etodos estad\'isticos de estimaci\'on de la confiabilidad del modelo hostil. Se realiza una comparaci\'on del rendimiento de tres m\'etodos estad\'isticos, tomando como referencia grafos utilizados previamente en la literatura especializada.

ABSTRACT. We present bi-objective optimization problems in partially reliable network design. Our models generalize the capacitated ring tree problem, which asks for a network that connects two types of given customers to a central node by combined ring and tree structures under capacity restrictions. Optional Steiner nodes are used as intermediate points if beneficial. The single-objective problem aims at minimizing the overall edge costs. We take into account the overall number of tree customers and edges as well as the maximal number of subtree customers and tree hops from leaf customers. Our new mathematical formulations for the bi-objective models are used to compute Pareto fronts within an exact iterative algorithmic framework. NP-hard subproblems are solved using branch and cut methods that incorporate problem specific valid inequalities. Objective space heuristics are embedded to speed up the algorithms which are tested on hard instances. We discuss related single-node failure scenarios.

ABSTRACT. Dentre as diversas ações realizadas na formação dos engenheiros, destacam-se as competições relacionadas à construção de aeromodelos de carga. O processo de construção de uma aeronave enquadra-se como um Projeto de Desenvolvimento de Produtos (PDP). Uma importante decisão na área envolve a seleção de pessoal, fator gerencial que afeta o desempenho do projeto. O presente trabalho tem a finalidade de propor um modelo de apoio à decisão multicritério para tomada de decisão em grupo quanto à seleção de pessoal para equipes de desenvolvimento de aeronaves para competição. O modelo é composto por duas etapas, onde a primeira envolve as avaliações individuais dos tomadores de decisão e a segunda agrega as opiniões dos envolvidos em uma opinião do grupo. O modelo proposto é aplicado para o caso da equipe Pegazuls Aerodesign, da Universidade Federal Rural do Semiárido, resultando na seleção adequada dos participantes que irão compor a equipe do projeto.

ABSTRACT. Personnel selection is an important decision that determines the competitiveness and improving of the performance of organizations. Personnel selection is a process which consists in to choose from among a set of candidates, those who possess the qualities required to perform a particular job. In this process, it is necessary to evaluate the performance of the candidates respect to a set of criteria, which may include subjective data. Personnel selection can be addressed as a multicriteria ranking problem, where at the end of the process, the best candidates should be at the top of the ranking. In this work, a real life application of personnel selection of 26 candidates in a software company is presented. For that, ELECTRE III method was used to obtain a valued outranking relation, which was exploited by using a multiobjective evolutionary algorithm to obtain a ranking of candidates. That algorithm tries to reduce the differences between the relation and the final ranking.

ABSTRACT. Los países en desarrollo necesitan atraer capitales que les permitan mejorar y ampliar su infraestructura y la calidad de la prestación de los servicios públicos con el propósito de fortalecer su desarrollo. Al momento de evaluar los préstamos e inversiones a estos países resulta importante contar con información que facilite diferenciar entre países con buena o mala calidad crediticia. En este trabajo se presenta un indicador del mérito al crédito de 98 países en desarrollo obtenido por aplicación del Método UTADIS (UTilités Additives DISciminantes) sobre la base de 7 criterios, que permite clasificarlos en 9 clases ordinales previamente determinadas empleando métodos estadísticos robustos. El modelo desarrollado fue capaz de clasificar correctamente el 91% de los países en su clase original y determinar el peso relativo de los 7 criterios empleados. Replicando la función de utilidad obtenida, el mérito al crédito de los países fue evaluado para el período 2008 – 2012.

ABSTRACT. El objetivo de la investigación es generar el diseño óptimo de rutas de los buses escolares para una red que describe el transporte de los colegios del norte de la ciudad de Bogotá, Colombia, incluyendo la secuencia de los buses en el nodo que representa el paso por la autopista (nodo que comunica los hogares de los estudiantes con los centros educativos) contribuyendo a la disminución de la congestión vehicular de esta zona ya que por la naturaleza de las calles este nodo representa un cuello de botella. Se definen dos modelos de programación entera mixta: rutas independientes (estudiantes de diferentes colegios no pueden compartir bus) y rutas compartidas (estudiantes de diferentes colegios pueden compartir bus). Estos modelos buscan minimizar el costo total de la operación utilizando una flota de buses homogénea que permita satisfacer los requerimientos de los colegios. Se presentan los resultados de ambos modelos y se realiza un análisis comparativo de los mismos.

ABSTRACT. En este trabajo se proponen un modelo y un método de solución al problema de configuración simultánea de rutas y frecuencias en un corredor troncal aislado de uno de los Sistemas Integrados de Transporte Masivo de Colombia (SITM-MIO). La formulación del problema, al igual que el método de solución, son adaptaciones de las presentadas en un trabajo existente, concebido para el servicio de transporte público de la ciudad de Tin Shui Wai, en Hong Kong. El modelo propuesto tiene en cuenta restricciones reales de la operación del SITM no encontrados en modelos previos. La propuesta se valida con datos reales de uno de los corredores más importantes del SITM-MIO. Los resultados obtenidos mejoran las soluciones actuales del sistema para dicho corredor.

ABSTRACT. In this research, a new variant of the vehicle routing problem is presented. The problem is inspired on disaster logistic applications, where the objective and some commercial constraints do not hold. In this variant, a set of geographically scattered required sites with known demands must be visited by a fleet of vehicles, while the objective function is the minimization of the sum of arrival times at required sites. This objective function better reflects the strategic goal under emergencies. In addition, sites can be visited by one or more vehicles in order to reach their demand. Note that avoiding split-delivery is a common constraint in commercial applications, while it can be preferred in humanitarian logistics. The problem is dealt by a metaheuristic approach based on the destroy and reconstruct concept. The algorithm is tested on classical benchmark instances for the CVRP available in the literature.

ABSTRACT. Se estudia el problema de localización de depósitos integrando las decisiones de ruteo y selección de flota de vehículos. Se presenta un modelo matemático basado en programación entera mixta, el cual supone que existe un conjunto discreto de depósitos candidatos con un costo de apertura fijo y capacidad determinada. Se considera una flota de vehículos heterogénea con capacidad limitada, y costos de utilización diferenciados por tipo de vehículo. Se asume una demanda determinística y un único periodo de planeación. Se propone un método de solución de tipo metaheurístico, que toma como base una búsqueda local iterativa. Se realizan pruebas computacionales en dos grupos de instancias tomadas y adaptadas de la literatura, que permiten ampliar la escasa literatura científica en la temática. Los resultados muestran la importancia de optimizar simultáneamente las decisiones de localización considerando costos de ruteo y selección de flota de vehículos.

ABSTRACT. The Capacitated Location Routing Problem (CLRP) is the combination of two well studied problems: the facility location problem and the capacitated vehicle routing problem. The objective is to minimize the cost of the open depots, the fixed cost of the vehicles and the cost of the routes. We present here a new variant, the Prize-Collecting Capacitated Location Routing Problem (PC-CLRP).In this case it is possible to leave customers unvisited and if a customer is visited it gives a gain. We proposed three integer linear programming formulations for the PC-CLRP (two-index and three-index flow models). We implemented a Branch-and-Cut algorithm based on one of them. Valid inequalities for the CLRP were adapted for the PC-CLRP. Also new valid inequalities and optimality cuts were proposed, together with their separation algorithms. A hierarchical branching strategy was also included. The initial solution was provided by an Ant Colony Algorithm. Computational results showed very promising.

ABSTRACT. Cuando se aplica la estrategia de descomposición cruzada al problema de localización se explota simultáneamente su estructura primal y dual, lo que resulta en la solución sucesiva de dos subproblemas y en la reducción del número de problemas maestros a resolver. Además, si se incorpora la relajación Lagrangeana Separable en este esquema cruzado, se garantiza que los subproblemas mantendrán todas las restricciones originales, se elimina la necesidad de resolver repetidamente alguno de los problemas maestros y con la estimación de los multiplicadores de forma heurística considerando el costo fijo y la demanda atendida en cada iteración, se obtiene un proceso de solución más eficiente y una estrategia más simple al interactuar entre dos subproblemas de transporte.

ABSTRACT. Urban logistics and last-mile distribution are a fundamental process in everyday activities for any city. Among alternatives to improve urban logistics operations, a very attractive one are loading and unloading bays, which have proven to be cost-effective solution. The process of designing an urban-transshipment network begins with establishing the location of the loading and unloading bays. First, we propose a methodology composed of a two sequential model scheme: (i) the quantity of bays needed is established using a queueing theory model; and then (ii) the specific location of bays is established with a location model based on a gravity approach. In the location analysis the approach considers demand characteristics and also others attributes and constraints of the area of analysis. Finally, an hypercube queueing model is used to determinate how the final proposal will respond to the usual activities. The proposed method is then tested in a case study in Santiago`s CBD.

ABSTRACT. En este trabajo se estudia un problema de localización de instalaciones con preferencias de los clientes, el cual es formulado con un modelo de programación binivel. Consideramos una restricción de capacidad para cada instalación donde cada una de ellas tiene un límite de clientes que se le puede asignar. La complejidad del problema radica en la existencia de soluciones múltiples en la asignación de los clientes a sus instalaciones preferidas cuando un número predefinido de instalaciones se ha localizado. Debido a que se quiere evitar resolver otro problema de optimización para solucionar el problema bajo el enfoque optimista, se aplica un método lexicográfico que garantiza la mejor conveniencia para el líder con una solución óptima del seguidor. Se propone un algoritmo genético para la obtención de buenas soluciones en un tiempo de cómputo razonable. Experimentación numérica se lleva a cabo mostrando la eficiencia del algoritmo propuesto para resolver la versión binivel del problema.

ABSTRACT. La ciudad de Punta Arenas se encuentra en la Patagonia chilena, dada su ubicación y características geográficas del lugar el abastecimiento de la ciudad y alrededores depende casi en su totalidad de los productos provenientes de la región central de Chile y de Argentina, donde en un 58% la carga es realizada utilizando medios terrestres. Dado el crecimiento de la ciudad y debido al impacto que tienen los camiones en las vías de tránsito en este año entra en vigencia una nueva restricción vehicular que impide el ingreso de camiones de gran tonelaje a la zona urbana de la ciudad. Debido a esto la ciudad requiere el desarrollo de un centro logístico fuera del límite urbano para la gestión y distribución de los productos, de forma tal de respetar la ley y satisfacer la demanda. En este artículo se presenta una propuesta de ubicación del centro logístico y el establecimiento de los posibles servicios a brindar por medio del uso de simulación, modelamiento matemático y optimización.

ABSTRACT. La mayor parte de la literatura sobre teoría de localización está enfocada a la ubicación de instalaciones deseables. Sin embargo, hay instalaciones que pueden ser molestas (no deseadas), pero son necesarias para el desarrollo de las comunidades. Esta molestia se vuelve una oposición para la localización de estos servicios. Los costos de instalación, transporte y operación, así como las emisiones contaminantes y la oposición de la población a la apertura de las instalaciones, hacen que las decisiones se tomen al resolver un problema con diversos objetivos. Estos aspectos pueden ser modelados considerando una estructura multiobjetivo y binivel. En este trabajo se presenta una propuesta de modelado para problemas de localización de instalaciones no deseadas en dos niveles con múltiples objetivos en el problema líder. Se considera el problema de mediana no deseada para la formulación y se concluye con una estrategia de solución desarrollada con base en la programación paramétrica.

ABSTRACT. O objetivo deste artigo é identificar e caracterizar os municípios eficientes em relação aos gastos públicos em educação nos municípios do estado do Rio de Janeiro utilizando a Análise Envoltória de Dados (DEA). Os dados foram coletados a partir de informações oficiais do Tesouro Nacional e do Ministério da Educação e foram divididos em duas partes: inputs (gastos municipais com Ensino Fundamental, número de professores do 9º ano do Ensino Fundamental e número de professores do 9º ano do Ensino Fundamental) e outputs (escalas de proficiência em português, escalas de proficiência em matemática e taxa de aprovação no 9º ano do Ensino Fundamental). O modelo utilizado foi o BCC orientado a output. Os resultados do artigo sugerem que a relação próxima entre a família e a comunidade escolar pode facilitar o aprendizado dos alunos. E por fim, conclui-se que o salário médio do professor pode ser um fator importante de motivação.

ABSTRACT. There is a vast literature referring to the development of efficient questionnaire designs for compensatory choice models. In compensatory models, consumers' utility function has a linear form and the parameters are estimated via the optimization of some real-valued function of the design's information matrix. However, the results are scarce with respect to non-compensatory models. In this work we study and develop efficient designs for the Elimination by Aspects (EBA) model, aiming at the elicitation of consumer's preferences using this decision rule to evaluate product alternatives in a choice-based conjoint context. We study existent techniques and develop new procedures to obtain the designs.

ABSTRACT. Higher Education Institutions (HEI) face the need to manage effectively and efficiently their resources because the demand for quality educational services from the private sector, government, and society. So as to measure the technical efficiency of 40 Mexicans HEI, we applied the data envelopment analysis (DEA) which is a methodology to empirically measure the productive efficiency of decision--making units (DMU). In this paper, we measure three core functions of every HEI: teaching, research, and knowledge dissemination. We particularly applied the output--oriented BCC-Primal model given that the CCR model does not accommodate returns to scales. As a result, the HEI are classified as either technical efficiency or technical inefficiency; thus, we identified potential improvements for the technical inefficient HEI. These results could support the decision--making process to improve the performance of the HEI.

ABSTRACT. This paper proposes the use of a Network DEA (NDEA) model to evaluate the efficiency of 16 professional teams of Defense of the Ancients 2 (known as DotA 2). The data used in this paper is from the World Championship organized by Valve in August 2015 at the city of Seattle. The NDEA approach in opposition to the standard DEA is more suitable to the configuration of the teams since it performs divisional interactions, the estimated efficiency produced is more representative of the process. Besides the efficiency index, which is used to establish a ranking, both efficiency and inefficiency sources among the teams separated are identified and analyzed. Subsequently, with the aims to reduce the benevolent nature of DEA models, the inverted frontier is used and a composed efficiency index is calculated taking into account both NDEA efficiency indexes and the inverted frontier.

ABSTRACT. El trabajo se enfoca en mostrar las ventajas que trae el modelo “Assurence Region Global” al momento de encontrar variables críticas que generan resultados de eficiencia poco confiables debido a la ausencia de dichas variables en el cálculo de la productividad y eficiencia de la DMU (Decision Making Unit) asociada. Para ello, se aplicó la técnica de Análisis Envolvente de Datos en una empresa del sector lácteo avaluando la eficiencia de sus productos principales. En primera instancia se aplicaron los modelos CCR-I y BCC-I para comparar el comportamiento de sus pesos ponderadores. En los resultados respectivos se obtuvieron valores de cero en dichos pesos, lo que generó ausencia de las variables asociadas a estos pesos y eficiencias poco confiables. Seguidamente, se implementó el modelo ARG-I-V para evidenciar el cambio en el desarrollo matemático y así, lograr valores de eficiencia confiables y equitativos para todas las DMU

ABSTRACT. Se presenta un método para cuantificar el impacto de la información pública en el precio accionario, usando para ellos redes neuronales artificiales y minería de texto. Para esto se recolecto una serie de noticias relacionadas al título accionario, para luego usar la minería de texto para agrupar y clasificar las noticias según patrones comunes que permitan asociarlas a priori un efecto común sobre la tendencia accionaria. Posteriormente se realiza una serie de experimentos, empleando modelos de redes neuronales para medir el impacto de las noticias. Los resultados muestran que al incorporar las noticias como una variable adicional al modelo de redes neuronales, aumenta en forma significativa la raíz del error cuadrático medio entre un 3,5% a 4,1% y el error porcentual absoluto medio aumenta entre un 5,2% a un 4,9 %. Finalmente, este trabajo contribuye entregando una evidencia experimental sobre un método de procesamiento de la información pública poco común y sobre su impacto en los resultados obtenidos en cuanto a la capacidad de un modelo de red neuronal para predecir el valor puntual y su tendencia.

ABSTRACT. En el presente trabajo se busca modelar el proceso de toma de decisiones de un inversor que debe elegir como componer su cartera de inversiones, entre un conjunto de acciones de once empresas, que son las que constituyen el índice MERVAL. Considerando que el objetivo del inversor es maximizar su rentabilidad, esta decisión se encuentra sometida a la incertidumbre del precio futuro de las acciones. Por ello se utiliza la metodología de las decisiones robustas, que busca obtener decisiones que, en lugar de maximizar u optimizar un valor esperado, se comporten lo mejor posible a lo largo de un espectro amplio de futuros probables.

ABSTRACT. This study analyzes the use of induced aggregation operators in Markowitz’s mean-variance portfolio selection model. The main advantage of this framework is that we can deal with complex attitudinal characters of the decision maker in the portfolio selection process. This may occur due to a wide range of circumstances including time pressure and competitiveness. The focus of the analysis is to replace the mean and the variance by the induced ordered weighted average and the induced ordered weighted average variance. Thus, instead of dealing with the information in a neutral way, the decision maker may use an induced attitude in the analysis. Some further extensions by using generalized aggregation operators are also considered. The work also presents an illustrative example in a financial decision analysis problem.

ABSTRACT. Triggered emails correspond to personalized messages sent automatically as a response to specific actions or states of the customers. Typical examples of this kind of campaigns include confirmation and order status emails, cross-selling recommendations, and re-engagement emails among others. This paper investigates the impact of triggered email marketing campaigns using an experimental approach, where automated emails were sent to a sample of 11.611 customers who had recently browsed the website of a multi-channel retailer but abandoned the process before purchasing. Our results show that browse abandonment triggered emails have significantly greater open rates, click through rates and incremental sales than traditional emails. In term of the design, contacting customers more than once and more personalized content positively influence the profitability of a campaign.

ABSTRACT. El sistema de educación superior chileno se encuentra enmarcado en un modelo competitivo, donde todas las instituciones generan acciones para captar un mayor y mejor número de postulantes. Se generaron tres modelos con variables explicativas, dos de ellos Logit, uno base y otro en el cual se incluía interacción entre las variables de entrada introducidas al modelo y un Árbol de Decisión, que contenía el mismo set de variables que el modelo Logit base. Los resultados que entregaron los modelos son que las ayudas económicas sí son relevantes en los alumnos y que afectan en distintos grados según el tipo de postulante. La demanda pasada de la carrera también es un factor importante, siendo el que más impacta en la decisión, como también lo son las actividades que realiza la universidad para generar conciencia de marca en sus postulantes. El modelo que tuvo mejores resultados en cuanto a la predicción de matrícula fue el Logit base, con un error promedio en la conversión de 5,7%.

ABSTRACT. Sales of ice-cream are one of the most weather-sensitive. With climate change, the periods of abnormal weather become frequent. We propose a model to measure precisely the impact of abnormal weather on ice-cream sales throughout the year. We show how using such a model, managers can devise mitigating strategies involving marketing, logistic, and financial measures. We model the best correlation between one or various weather variable and consumer buying patterns in the corresponding regions. The method even enables the modeling of ice-cream sales in winter with the necessary precision so as to allow managers to understand sales performance even at Christmas. We present a new methodology to help managers in all fast selling consumer goods. Models developed with our methodology allow managers to better understand the role of weather on business performance, take real-time corrective actions, and protect sales as well as marketing efforts from the effects of adverse weather.

ABSTRACT. Esta paper propone y genera un prototipo de sistema de recomendación, que permite ofrecer a los clientes de un sitio de cupones online, las emails con productos ofertados que más se ajustan a sus necesidades, y lograr así mejorar las conversiones y los ingresos de la empresa. Se generaron dos modelos principales para la recomendación, uno basado en filtros colaborativos sobre la base de los usuarios y otro basado en reglas de asociación. Los resultados de la experimentación arrojaron que la Tasa de Clicks y la Tasa de Conversión (porcentaje de compras del total de clics) aumentan utilizando un modelo de recomendación con un 95% de confianza. El modelo que mejores resultados obtuvo, es el de Reglas de Asociación con Productos Frecuentes. Extrapolando las cifras a toda la base de datos de la empresa, el correo tendría un tráfico diario adicional de aproximadamente 4.500 clientes, 1.000 adicionales en su sitio web y su Tasa de Conversión aumentaría en un 40% en promedio.

ABSTRACT. Methods for individually adapted choice-based conjoint analysis requires a precise quantification of the uncertainty associated with preference parameters. Bayesian approaches require impractical enumeration for adaptive selection of questions. Polyhedral methods, which quantify the uncertainty through a (convex) polyhedral set, allow for quick updates and effective optimization-based question selection. However, because it does not consider respondent errors explicitly, its performance can deteriorate with high error rates. In this talk we show how replacing the sets used in the polyhedral approach with ellipsoids we can include respondent error and develop a purely geometric approach that is as intuitive as the polyhedral approach, but nearly matches what a Bayesian approach would do. This ellipsoidal approach extends the effectiveness of the polyhedral approach to the high heterogeneity and high response error setting and provides a geometric interpretation of Bayesian approaches.