ENBIS-16 in Sheffield

11 – 15 September 2016; Sheffield Abstract submission: 20 March – 4 July 2016

My abstracts

 

The following abstracts have been accepted for this event:

  • Tree-Based Methods for (Life) Insurance and Some Considerations about Statistics and Actuarial Mathematics

    Authors: Walter Olbricht (University of Bayreuth)
    Primary area of focus / application: Business
    Secondary area of focus / application: Mining
    Keywords: Tree-based methods, Life insurance, Specific data features, Internal structure of the data
    Submitted at 28-May-2016 13:03 by Walter Olbricht
    Accepted
    14-Sep-2016 10:30 Tree-Based Methods for (Life) Insurance and Some Considerations about Statistics and Actuarial Mathematics
    Statistics and actuarial mathematics share a lot of common ground. However, there are also obvious differences. Data sets in the field of insurance tend to be very large so that sampling aspects and random errors are not of prime concern. On the other hand they are typically heterogeneous so that substructures matter. Furthermore, frequently contextual knowledge (such as shifts in legislation which had effects in the data) is available which could - and should - be incorporated. Similar features seem to occur in many business applications.
    The talk analyses this background and suggests tree-based methods as an interesting statistical tool even for the classical field of life insurance. In particular a “hybrid” approach (using regression trees for a classification situation) is proposed. The main advantage of this approach is its ease of interpretability and its inherent transparency.
  • A Bayesian Self-Starting Method for Online Monitoring of Phase I Data

    Authors: Panagiotis Tsiamyrtzis (Athens University of Economics and Business), Konstantinos Bourazas (Athens University of Economics and Business), Dimitrios Kiagias (University of Sheffield)
    Primary area of focus / application: Process
    Secondary area of focus / application: Modelling
    Keywords: Statistical Process Control and monitoring, Predictive distribution, Exponential family, Phase I outlier detection
    Submitted at 29-May-2016 13:46 by Panagiotis Tsiamyrtzis
    Accepted (view paper)
    13-Sep-2016 16:00 A Bayesian Self-Starting Method for Online Monitoring of Phase I Data
    Standard Statistical Process Control/Monitoring (SPC/M) charts typically use a phase I/II split for calibration and online testing respectively. The phase I data assumed to fulfill certain standards (e.g. being a random sample from the in control distribution) and one can test them in an off-line mode (i.e. only once the phase I data collection is completed). Some frequentist self-starting methods attempt to improve things by performing calibration and testing simultaneously.
    In this work a Bayesian alternative is proposed, which utilizes the (usually) available prior distribution to provide a chart based on the predictive distribution of a future observable (Predictive Control Chart – PCC). It is self-starting and performs online monitoring, right after the first observable becomes available. PCC will be presented in its most general form, allowing data of any (discrete or continuous) distribution as long as it is a member of the regular exponential family. We will also establish that PCC generalizes frequentist self-starting methods. Simulations will examine its performance against the frequentist based methods and a real data set will illustrate its use.
  • Bayesian Design of Experiments via Gaussian Process Emulation

    Authors: David Woods (University of Southampton)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Computer experiments, Generalised linear models, High dimensional design, Nonlinear design, Smoothing
    Submitted at 30-May-2016 11:43 by David Woods
    Accepted
    12-Sep-2016 15:00 Bayesian Design of Experiments via Gaussian Process Emulation
    The design of any experiment is implicitly Bayesian, with prior knowledge being used informally to aid decisions such as which factors to vary and the choice of plausible causal relationships between the factors and measured responses. Bayesian methods allow uncertainty in these decisions to be incorporated into design selection through prior distributions that encapsulate information available from scientific knowledge or previous experimentation. Further, a design may be explicitly tailored to the aim of the experiment through a decision-theoretic approach with an appropriate loss function.

    When designing experiments for nonlinear parametric models, finding a Bayesian optimal design is typically analytically intractable and often computational infeasible. The expected loss usually involves an intractable and high dimensional integral. We will present methodology for mitigating the computational expense of design through combining an application of Gaussian Process (GP) regression models with a cyclic descent (coordinate exchange) optimisation algorithm. We adopt a methodology from the field of computer experiments and build a GP emulator for the expected loss. The methods allow optimal designs to be found for previously infeasible problems. We will describe the methodology and demonstrate it on a variety of examples.
  • Effects of Imperfect Testing on Presence-Absence Sampling Plans

    Authors: Edgar Santos-Fernández (Massey University), K. Govindaraju (Massey University), Geoff Jones (Massey University)
    Primary area of focus / application: Quality
    Secondary area of focus / application: Modelling
    Keywords: Sampling inspection plan, Risk assessment, Measurement errors, Presence-absence tests, Bayesian inference, Sensitivity and specificity
    Submitted at 30-May-2016 11:56 by Edgar Santos Fernandez
    Accepted (view paper)
    12-Sep-2016 10:00 Effects of Imperfect Testing on Presence-Absence Sampling Plans
    Sampling inspection plans are used in the food industry to determine whether a batch of food is contaminated or not. Testing for pathogens is mandatory in several foodstuffs because some pathogenic bacteria pose a significant risk for human health, even when these are consumed in minute quantity. Test performance measures such as sensitivity and specificity are generally ignored in microbiological risk assessment. In this presentation, we examine the impact of imperfect analytical tests on sampling inspection plans for presence-absence characteristics. We discuss several plausible scenarios and assess the risk for the consumers. The method is illustrated using collected data over two years for Cronobacter spp. in skimmed milk powder. The probability of contamination and the test sensitivity and specificity, are estimated using Bayesian inference. We examine the sampling plans proposed by the Codex Alimentarius and by New Zealand's Ministry of Primary Industries. A cost analysis is carried out to show the economic loss due to measurement errors. We describe the strengths and limitations of these inspection plans under different conditions and propose a plan that could provide better protection to the consumers as well as to the producers.
  • Bespoke Online App for Design of Experiments for a Research Robot

    Authors: Winfried Theis (Shell Global Solutions International B.V.), Mark Brewer (Shell Global Solutions International B.V.), Axel Makurat (Shell Global Solutions International B.V.)
    Primary area of focus / application: Design and analysis of experiments
    Secondary area of focus / application: Consulting
    Keywords: Design of Experiments, Mixture design, Space filling, Creating web applications, R Shiny
    Submitted at 30-May-2016 15:47 by Winfried Theis
    Accepted
    14-Sep-2016 09:00 Bespoke Online App for Design of Experiments for a Research Robot
    Especially when a robot is used for experimental work it becomes important to plan the experiments well, as otherwise a lot of data with not too much information might be produced. We were approached by colleagues who just had started using a mixing robot for experimental work. The group is working on chemical mixtures that need to be tailor-made for each application of the basic technology a lot of testing is required in more of an exploratory fashion. Therefore we chose to implement for them a R-Shiny application to create space-filling designs that respect the mixture requirements and some other specialities of the method. It also needed to have an output that allows easy input of the design into the mixing robot steering program.
    In this talk we will explain the way how we solved the various potential restrictions on the design space but also will show how simple it was to create an online user interface with R-Shiny and demo it shortly.
  • A Bayesian Model for Count Data Affected by False Positives and False Negatives

    Authors: Francesca Pennecchi (Istituto Nazionale di Ricerca Metrologica - INRIM), Walter Bich (Istituto Nazionale di Ricerca Metrologica - INRIM), Cinzia Carota (Dipartimento di Economia e Statistica “Cognetti de Martiis”, Università degli Studi di Torino), Giancarlo D’Agostino (Istituto Nazionale di Ricerca Metrologica - INRIM), Marco Di Luzio (Istituto Nazionale di Ricerca Metrologica - INRIM), Consuelo R. Nava (Dipartimento di Economia e Statistica “Cognetti de Martiis”, Università degli Studi di Torino), Alessio Ricci (Dipartimento di Economia e Statistica “Cognetti de Martiis”, Università degli Studi di Torino)
    Primary area of focus / application: Metrology & measurement systems analysis
    Secondary area of focus / application: Modelling
    Keywords: Count data, Bayesian inference, False positives, False negatives, Measurement uncertainty
    Submitted at 30-May-2016 16:48 by Francesca Pennecchi
    Accepted
    Count data are very common in many scientific and technological fields such as optics, ionizing radiations, microbiology and chemistry, as well as in many everyday activities. However, within the metrological community little attention has been given to measurements based on counting discrete quantities. As for any other measurement, the result of a counting procedure should consist of an estimate of the unknown number of items and an associated uncertainty. Recently, a general model for this type of measurements has been proposed to the metrological community in order to estimate the unknown count and to evaluate its uncertainty in a way compliant with the general framework of the “Evaluation of measurement data – Guide to the expression of uncertainty in measurement”.
    In the present work, we address the same problem by using a hierarchical Bayesian model, which takes into account the probabilities of false positives and false negatives, inducing overestimates and underestimates of the measurand, i.e. the unknown count, respectively. Prior information on the measurand is introduced at the first level of the hierarchy, then we enrich the model by adding a second level where further information on such two probabilities is exploited.
    Such a model is applied to a radioactivity measurement problem where the measurand is the number of radioactive atoms decaying in a definite counting time. Due to the dead time of the detection system - a stand-by period after the counting of each event which follows gamma photon detection and is required to process it - there is a non-zero probability that some events are not counted. The posterior probability density function of the measurand proves to be able to correct for the presence of such false negatives, providing a reliable estimate of the number of decayed atoms.