Catálogo de publicaciones - libros

Support Vector Machines (SVM) from statistical learning are very powerful methods which can be used as (e.g.binary) classifiers or discriminators in a wide range of applications. Advantages of SVM are that weak prior assumptions about both model and data suffice. Moreover, optimization of the SVM essentially regularizes the emerging data model by restricting the model to special data points, the support vectors, usually a small subset from the training data. In our paper we discuss ways of detecting informative and typical subsets from SVM solutions, with the aim of extracting simple rules.

We present a model for the pressure between positions in a network. This notion is made operational through identifying the locations and sources of network value. Precisely these ‘markets’, where network value is obtained and generated, can be at stake and determine the pressure a position may experience. Using a limit approach, we also derive our square law of network power.

In this paper, we describe an extension of the OLAP (On-Line Analytical Processing) framework with automated causal diagnosis, offering the possibility to automatically generate explanations and diagnostics to support business decision tasks. This functionality can be provided by extending the conventional OLAP system with an explanation formalism, which mimics the work of business decision makers in diagnostic processes. The central goal of this paper is the identification of specific knowledge structures and reasoning methods required to construct computerized explanations from multidimensional data and business models. The methodology was tested on a case study involving the comparison of financial results of a firm’s business units.

Several business intelligence concepts show that it is possible to build a data warehouse based on heterogeneous quantitative and qualitative data. But the danger of flooding the management with information still remains. Business information should be made available according to the personal need of a manager by a self-defined push mechanism. The concept of active data warehousing is going to be expanded in this way, first. Messages about new relevant information should be created for quantitative and qualitative data as well. Due to this, there is a need for an enterprise specific ontology. This ontology works as an intermediate between current information and user profiles. Just interesting information are passed to the user. The second approach clusters qualitative data in favour of visualization. The usage of increasing cell structures provides a self organizing map. Decision makers will get the opportunity to search in document volumes, which are clustered.

Im Rahmen der High-Tech-Offensive Bayern (HTO) wurde an der Fachhochschule Augsburg für die Virtuelle Hochschule Bayern (VHB) eine Web-basierte „Selbstleraumgebung Operations Research (OR)“ entwickelt. Ziel war es, vornehmlich Studierenden der Fachrichtungen Wirtschaftsinformatik und Informatik, eine OR-Vorlesung virtuell bzw. online anzubieten, mit der sie sich die grundlegenden OR-Verfahren im Selbststudium aneignen können.

Das unter diesen Voraussetzungen entwickelte Lernprogramm wird seit gut einem Jahr intensiv von Studierenden verschiedener Hochschulen genutzt. Wesentliche Besonderheit dieses Lernprogramms ist der explorative Charakter:

Neben dem eigentlichen Lehrbuch, das im Wesentlichen aus einem Online-Skriptum besteht, gibt es eine Sammlung von über 60 Java-Applets. Diese simulieren die theoretisch vorgestellten Algorithmen anhand von frei wählbaren Werten, indem sie jeden Schritt erklären. In diesen Simulationen kann der Lernende schrittweise die Algorithmen an eigenen Beispielen selbst „entdecken“.

Das umgesetzte Konzept ermöglicht den Studenten auch einen iterativen Lernprozess und die Möglichkeit einer effizienten Wissensauffrischung. Die bisherigen Erfahrungen mit dem Lernprogramm und die durchgeführten Studentenevaluationen werden vorgestellt und grob analysiert.

Die Selbstlernumgebung im Gesamten entspricht den Vorstellungen der deutschen HRK (Hochschulrektorenkonferenz), die in Selbstlernumgebungen Lehrangebote sieht, die eigenständig und selbstbestimmend und praktisch ohne Betreuung von den Lernenden bearbeitet werden und die damit auf die Lehrbedürfnisse, Lerngeschwindigkeiten und das individuelle Zeitbudget Rücksicht nehmen.

Eine weitere Besonderheit ist die einfache Erweiterbarkeit dieses Web-basierten Lernprogramms. Die flache Struktur erlaubt es weitere Texte und Applets zu integrieren. Um eine einfache Erstellung von interaktiven Übungen zu ermöglichen, wurde ein Autorenwerkzeug entwickelt, das es ermöglicht, ohne jegliche Java Kenntnisse, Java-Übungs-Applets mit automatischer Auswertung zu erstellen. Zum Abschluss des Vortrags wird dieses Werkzeug, das mittlerweile auch in anderen E-Learning-Projekten eingesetzt wird, in seinen elementaren Funktionen vorgestellt.

Crew scheduling for airlines requires an optimally scheduled coverage of flights with regard to given timetables. We consider the crew scheduling and rostering process for airlines, where crew members are stationed unevenly among home bases. In addition, their availability changes dynamically during the planning period due to pre-scheduled activities, such as office and simulator duties, vacancy, or requested off-duty days.

Besides this highly complex setting, crew satisfaction becomes more and more important: Recent approaches focus on a variety of quality-of-life criteria like the consideration of individual preferences. Hence most of them are applied to individual rosters; certain aspects especially for the appropriate handling of teams are neglected. Therefore the introduced concept of Team-oriented Rostering addresses the upcoming demand for a reduction of the usually high amount of team changes among crew members within daily or day-by-day team compositions. It relies on the optimal set of roster combinations for the scheduled crew members within the assignment step.

The combination of process integration and Operations Research enables an integrated technique assessment and a subsequent process design. The application of Multi Objective Pinch Analysis (MOPA) permits the identification of overall saving potentials for energy, water and Volatile Organic Compounds (VOC). In this paper the general concept of MOPA is described and its application is shown in its first steps for a bicycle coating plant in Chile.

Manufacturing companies in a make-to-order environment sell customer specific products. Usually, a company offers more than one product, faces a stochastic demand, and a time-varying product price. The latter cannot entirely be set autonomously by the company due to competitors. The decision the company has to make for an incoming order is whether to accept or reject the order, depending on the remaining capacity, the contribution margin of the order and the orders expected for the future. In our contribution we will discuss if the methods for the airline revenue management are applicable to the described problem. After giving an overview about the requirements a decision support system ought to fulfill to improve the order selection process in a make-to-order environment, we will discuss the differences of such a system to existing revenue management approaches in service industries. Subsequently, we give a mathematical formulation for the described problem and apply it to a simplified practical example.

In this paper the approach of combining predictions is used to benefit from the advantages of forecasts predicting on different levels, to reduce the risks of high noise terms on low level predictions and overgeneralization on higher levels. The presented experimentally compared approaches of combining seasonal airline demand forecasts differ concerning input decomposition, multilevel structures, combination models and kinds of aggregation. Significant forecast improvements have been obtained when using multilevel, hierarchical structures.