ENBIS-8 in Athens

21 – 25 September 2008 Abstract submission: 14 March – 11 August 2008

My abstracts

 

The following abstracts have been accepted for this event:

  • Comparing different transformation strategies on a six-factor full factorial (2^6) industrial experiment.

    Authors: Erik Mønness
    Affiliation: Hedmark University College
    Primary area of focus / application:
    Submitted at 30-Apr-2008 08:01 by Erik Monness
    Accepted (view paper)
    22-Sep-2008 16:00 Comparing different transformation strategies on a six-factor full factorial (2^6) industrial experiment.
    Different transformation strategies are compared on a data set from an experiment to determine the optimum operating conditions for a milling machine with respect to surface finish. The data was obtained from a six-factor full factorial (2^6) Designed Experiment. The six factors Tool speed, Work piece speed, Depth of cut, Coolant, Direction of cut and Number of cuts were varied each with two levels. Each treatment was repeated 8 times so the mean and standard deviation of each treatment were available. (A talk on comparing fractions from this experiment was given at ENBIS-2005, and published in Applied Stochastic Models in Business and Industry, 2007. 23: p. 117-128: Comparing different fractions of a factorial design: A metal cutting case study. Mønness, Linsley, and Garzon,.).


    An empirical power transformation is determined graphically from the σ≈μ^(λ-1)relation. Another strategy is derived from the Box-Cox transformation.

    Issues to be explored:
    • What λ is best based on the saturated model?
    • Is the estimate stable considering different fractions of the data?
    Models involving the experimental factors:
    • Applying different models, e.g. involving only main effects, 2-interactions, 3-interactions, how stable is the λ estimate?
    • How does the transformation affect the set of significant factors?
    Since one motivation for doing a transformation is to simplify any model, this issue is of interest when having many experimental factors.
  • Cumulative copula charts for controlling the dependence among multivariate observations

    Authors: András Zempléni, Pál Rakonczai, Csilla Hajas
    Affiliation: Eötvös Loránd University, Budapest
    Primary area of focus / application:
    Submitted at 30-Apr-2008 08:08 by András Zempléni
    Accepted (view paper)
    22-Sep-2008 15:20 Cumulative copula charts for controlling the dependence among multivariate observations
    Dependence plays an important role in the properties of multivariate observation. In order to monitor possible changes in this aspect, beyond the usual approach of covariance estimation, we propose an additional control chart, which is sensitive to departures from the assumed dependence structure. This chart is cumulative in the sense that at every step all the available data to that time point is utilized.
    We use the popular copula-approach for this approach. The copula allows for investigation the dependence structure separately from the univariate distributions, which is especially useful if the marginal distributions may change without causing an error, like in our case. Assuming that the standard charts deal with the univariate modeling as well as the changes in the mean vector, with the proposed new chart all aspects of important, possible changes are taken care for.
    The fit of the observed copula to the one, considered as nullhypothesis can be measured by the multivariate version of the probability integral transform. Critical values can be constructed by simulation (see Rakonczai and A. Zempléni, 2007 for example). We illustrate the use of the chart by simulation studies and by real life bivariate data from our university, where the dependence between high-school final examination results and average achievements at the university is investigated.

    Reference

    P. Rakonczai and A. Zempléni, 2007: „ Copulas and goodness of fit tests " in: Recent Advances in Stochastic Modelling and Data Analysis, World Scientific, (Editor: C. Skiadas)
  • The use of factor and cluster analysis in marketing research.

    Authors: Antigone G. Kyrousi, Athanasios G. Poulis
    Primary area of focus / application:
    Submitted at 30-Apr-2008 11:07 by Athanasios Poulis
    Accepted
    23-Sep-2008 09:40 The use of factor and cluster analysis in marketing research.
    Factor and cluster analysis have become common tools for the marketing researcher. Both of them are the most frequently employed methods of research in a marketing context.

    In this paper, a distinction is made between exploratory and confirmatory factor analysis while the existing literature is being reviewed. Recent trends, such as the increasing emphasis placed upon confirmatory rather than exploratory factor analysis, as well as the use of factor analysis along with conjoint analysis, are further identified.

    In addition, cluster analysis, which is a set of methodologies, is being identified as the most common tool for classification in the marketing field. However, its problematic use especially in qualitative research has been observed and thus, suggestions have been proposed for the marketing researcher.

    Furthermore, some recent applications of factor and cluster analysis in marketing problems are being displayed in order to illustrate the problems at hand. Finally, conclusions are drawn, regarding the misapplications of cluster and factor analysis in marketing research.
  • Iterative Designs of experiments for constraint approximation

    Authors: Picheny V., Ginsbourger D., Roustant O., Haftka R.T.
    Affiliation: Ecole Nationale Superieure des Mines de Saint Etienne / University of Florida
    Primary area of focus / application:
    Submitted at 30-Apr-2008 14:10 by Victor Picheny
    Accepted
    23-Sep-2008 15:20 Iterative Designs of experiments for constraint approximation
    This article presents a new criterion for design of experiments, when a metamodel (Kriging) is used to approximate a function that must be known accurately at a particular level-set. Such context occurs for instance in surrogate-based optimization when the constraint function is approximated by a surrogate, or in propagation of uncertainty, when a surrogate is used to compute a probability of failure.
    We propose a modification of the classical IMSE criterion, based on an explicit trade-off between reduction of global uncertainty and exploration of target regions, by using the statistical information given by the Kriging model. Sequential strategies are then used to build optimal designs of experiments.
    The method is illustrated on several test-problems of dimensions one, two and six. It is shown that compared to classical space-filling strategies, the error on target regions can be reduced very significantly, with reasonable pay-off on the global accuracy.
    Finally, the method is tested on a propagation of uncertainty problem, resulting with a gain in accuracy of several orders of magnitude on the probability of failure compared to space-filling designs.
  • A Methodology for the Detection of Segments of Clients with Homogenous Water Consumption Habits. Application to the City of Barcelona.

    Authors: S. Fontdecaba, L. Marco, V. Martinez de Pablo, L. Rodero, J.A. Sanchez,I. Sole, X. Tort-Martorell, J. Zubelzu
    Affiliation: UPC (Universitat Politècnica de Catalunya) and Grupo AGBAR
    Primary area of focus / application:
    Submitted at 30-Apr-2008 16:52 by Xavier Tort-Martorell
    Accepted
    22-Sep-2008 10:55 A Methodology for the Detection of Segments of Clients with Homogenous Water Consumption Habits. Application to the City of Barcelona.
    Water management has become a vital concern for both water supply companies and public administrations due to its importance for life and current scarcity in many areas. Many studies have been made trying to explain which factors influence water demand. These studies commonly use ordinary least squares to explain domestic water consumption mainly using economical variables (price), socio-demographic variables (population, age structure of population, nationality) and territorial variables (population density).
    Prior to having models for predicting water demand, the goal of our approach has been segmenting clients into meaningful groups in order to better understand the water consumption behaviour. The study is based on a database with more than one million observations from Barcelona and surrounding towns.
    Cluster analysis and decision trees have been used to obtain a descriptive segmentation of clients. The result is a stable partition in 6 groups. This segmentation is the same with and without considering the actual water consumption as a variable, which entails a different behaviour on water consumption depending on socio-economical variables. Although the study is based on the case of the Barcelona area, it allows the development of a general methodology, which will be described in the presentation.
  • Optimal designs for Gaussian random field regression models

    Authors: Maroussa Zagoraiou, Alessandro Baldi Antognini
    Affiliation: Department of Statistical Sciences, University of Bologna, Italy
    Primary area of focus / application:
    Submitted at 30-Apr-2008 16:57 by Alessandro Baldi Antognini
    Accepted (view paper)
    24-Sep-2008 09:20 Optimal designs for Gaussian random field regression models
    The present paper deals with optimal designs for random field regression models in the univariate case.
    In particular we consider the problem of designing experiments (i.e. sampling in time or in the real line) when the observations can be modelled via a Gaussian process with a regressive trend component and an exponential correlation structure.
    This modelling approach has been widely used for the analysis of computer experiments
    (see for instance Sacks, Welch, Mitchell and Wynn, 1989) and in empirical and theoretical finance, in order to model continuous time interest rates (see Gourieroux and Jasiak, 2008).
    Assuming the Maximum Likelihood approach, we study the optimal design problem for the estimation of the unknown parameters of the model using a criterion based on the Fisher information matrix.

    Sacks J., Welch W.J., Mitchell T.J. and Wynn H.P. (1989) Design and analysis of computer experiments, Statistical Sciences, 4, 409-423.
    Gourieroux C. and Jasiak J. (2008) Financial Econometrics: Problems, Models, and Methods. Princeton University Press