Showing papers in "Quality Technology and Quantitative Management in 2014"

Efficient Construction of Decision Trees by the Dual Information Distance Method

Abstract: The construction of efficient decision and classification trees is a fundamental task in Big Data analytics which is known to be NP-hard. Accordingly, many greedy heuristics were suggested for the construction of decision-trees, but were found to result in local-optimum solutions. In this work we present the dual information distance (DID) method for efficient construction of decision trees that is computationally attractive, yet relatively robust to noise. The DID heuristic selects features by considering both their immediate contribution to the classification, as well as their future potential effects. It represents the construction of classification trees by finding the shortest paths over a graph of partitions that are defined by the selected features. The DID method takes into account both the orthogonality between the selected partitions, as well as the reduction of uncertainty on the class partition given the selected attributes. We show that the DID method often outperforms popular classifiers, in terms of average depth and classification accuracy.

Causal Structure Learning and Inference: A Selective Review

Abstract: In this paper we give a review of recent causal inference methods. First, we discuss methods for causal structure learning from observational data when confounders are not present and have a close look at methods for exact identifiability. We then turn to methods which allow for a mix of observational and interventional data, where we also touch on active learning strategies. We also discuss methods which allow arbitrarily complex structures of hidden variables. Second, we present approaches for estimating the interventional distribution and causal effects given the (true or estimated) causal structure. We close with a note on available software and two examples on real data.

Official Statistics Data Integration Using Copulas

Abstract: The aim of this paper is to propose a novel approach to integrate financial information, incorporating the dependence structure among the variables in the model. The approach is based on two types of graphical models: vines and non-parametric Bayesian belief nets (NPBBNs). Vines are undirected graphs, representing pair copula constructions, which are used to model the dependence structure of a set of variables. NPBBNs are directed graphs, that use pair copulas to model the dependencies, and allow US for diagnosis and prediction via conditionalization. This approach permits to aggregate information and to calibrate the results obtained with different sources of data. The illustrated methodologies are applied to two financial datasets, the first one containing data collected through a survey and the second one containing official statistics data.

Optimal Decisions on the Accelerated Degradation Test Plan Under the Wiener Process

Abstract: The cumulative damage of highly reliable products that subject to multiple loading stresses is investigated under Wiener process. Optimal strategies on the constant-stress accelerated degradation test plan are established to reach a compromised decision between the experiment budget and the estimation precision on the reliability inference. An algorithm is provided to search an optimal strategy for the accelerated degradation test. An example of light emitting diodes is used for illustrating the application of the proposed method.

Maintenance Service Contracts for Equipment Sold with Two Dimensional Warranties

Abstract: HeThis paper deals with maintenance service contracts for equipment (such as dump trucks) sold with two dimensional warranties. We consider a situation where an agent offers several maintenance contract options and the owner of the equipment has to select the optimal option. As the availability of the equipment is a critical measure for attaining the owner’s business objectives, the maintenance contract options offered need to ensure a high availability of the equipment. We study three maintenance service options considering the availability target from both the owner and OEM point of views and use a non-cooperative game formulation to determine the optimal price structure (i.e., the contract price and repair cost) for the OEM and the optimal contract option for the owner.

Maintenance Versus Individual and Overarching Protections for Parallel Systems

Abstract: HeA parallel system consisting of components with different characteristics is studied. The components can be unavailable due to internal failures or external impacts. Each component has an increasing failure rate and is subjected to external impacts that can occur with fixed frequency. In order to increase the system availability, three measures can be taken: 1. Increase of the components replacement frequency; 2. Overarching protection of the system; 3. Individual component protection. The external impact must penetrate both the overarching protection and the individual component protection in order to destroy a component. The optimal resource allocation strategy which minimizes the total cost of maintenance, protection and damage caused by unsupplied demand is studied. The proposed approach is based on a universal generating function technique and a genetic algorithm.

Transient Analysis of Software Rejuvenation Policies in Virtualized System: Phase-Type Expansion Approach

Abstract: HeThis paper presents a transient analysis of software rejuvenation with virtualization. In particular, we apply the phase-type expansion to analyze the pointwise availability of virtual-machine (VM) based software rejuvenation with two policies; cold-VM and warm-VM rejuvenation policies. The performance measures can be derived from the stochastic models described by labeled Markov regenerative stochastic Petri Nets (MRSPNs).

A Failure-rate-reduction Periodic Preventive Maintenance Model with Delayed Initial Time in a Finite Time Period

Abstract: HeA new machine will not fail easily in the early stage of its useful life. This phenomenon is consistent with the fact that the optimal sequential preventive maintenance (PM) policy which has longer PM intervals in its earlier stage of lifetime. In this paper, we propose a failure-rate-reduction periodic PM model with delayed initial time in a finite time span. Then, the optimal periodic PM policy is developed by minimizing the expected total maintenance cost, which can have smaller expected total maintenance cost than the optimal policy of the original failure-rate-reduction periodic PM model. The algorithm of finding the optimal PM policy for the proposed PM model is developed. Finally, examples are illustrated to verify the optimal policies of the proposed new PM model.

Joint Reliability Importance in a Binary K-out-of-n: G System with Exchangeable Dependent Components

Abstract: In this paper, we study joint reliability importance (JRI) in a k-out-of-n : G structure consisting of exchangeable dependent components. We obtain a closed-form formula for the JRI of multiple com...

Change Point Estimation Based Statistical Monitoring with Variable Time Between Events (TBE) Control Charts

Abstract: Statistical process control efforts are considered to ensure high-quality production and reduce costs in the competitive environment of business. The signal issued by a control chart trigge...

Travellers' Satisfaction with Railway Transport: A Bayesian Network Approach

Abstract: Survey data are often employed as a tool to support planners in defining and reviewing their intervention programs and policies. In the current literature, a broad range of studies have examined the determinants and components of consumers' satisfaction. Most of these studies have exploited statistical models, such as probit or logit, to determine the categorical dependent variables. Less attention has been paid to the investigation of the structural properties of the data by testing alternative methods and procedures. The present work performs such an analysis using data from a survey on consumer satisfaction from railway transport in 14 EU countries. The data were collected in a large survey (more than 17,000 observations) conducted on behalf of the European Commission. By applying both ordered logistic regression and Bayesian Network (BN) analysis, the research question addressed by this paper is twofold. First of all, the performance of the two methodologies, defined in terms of their predictive capability, is assessed. Secondly, the main policy implications conveyed by the two models are compared. According to the results, BN analysis provides better predictions of the outcomes of customers' satisfaction with railway transport. Moreover, the choice of the statistical methodology is a relevant issue for policy-making, since the policy messages conveyed by the two models differ significantly.

Equilibrium balking strategies in the Geo/Geo/1 queues with server breakdowns and repairs

Abstract: This paper studies the equilibrium behavior of customers in the Geo/Geo/1 queue with server breakdowns and repairs. The reliability factor is introduced into the economic analysis and customers' behavior analysis of the discrete-time queues. The server is subject to breakdowns and repairs when providing service to customers. Upon arrival, the customers decide for themselves whether to enter the system or balk based on a natural reward-cost structure and available information on hand. The behavior of customers is investigated with various levels of information regarding three states of server and the system queue length. By deriving and solving a set of system difference equations, we obtain the stationary distribution of the system and the mean sojourn time of an arriving customer. Equilibrium strategies for the customers under different levels of information are derived and the behaviors of the customers under these strategies are investigated. Finally we illustrate the effect of several parameters on the equilibrium behavior via numerical experiments.

Inferring Networks from Multiple Samples with Consensus LASSO

Abstract: Networks are very useful tools to decipher complex regulatory relationships between genes in an organism. Most work address this issue in the context of i.i.d., treated vs. control or time-series samples. However, many data sets include expression obtained for the same cell type of an organism, but in several conditions. We introduce a novel method for inferring networks from samples obtained in various but related experimental conditions. This approach is based on a double penalization: a first penalty aims at controlling the global sparsity of the solution whilst a second penalty is used to make condition-specific networks consistent with a consensual network. This ''consensual network'' is introduced to represent the dependency structure between genes, which is shared by all conditions. We show that different ''consensus'' penalty can be used, some integrating prior (e.g., bibliographic) knowledge and others that are adapted along the optimization scheme. In all situations, the proposed double penalty can be expressed in terms of a LASSO problem and hence, solved using standard approaches which address quadratic problems with $L_1$-regularization. This approach is combined with a bootstrap approach and is made available in the R package therese. Our proposal is illustrated on simulated datasets and compared with independent estimations and alternative methods. It is also applied to a real dataset to emphasize the differences in regulatory networks before and after a low-calorie diet.

An Inspection Model Subject to Small Stoppages and Hard Failures

Abstract: HeThis paper considers an inspection problem of a production line subject to small stoppages and hard failures. We assume that the probability density function of time to hard failure is a function...

Enterprises' Treasury Management Strategy in Post Financial Crisis Era

Abstract: Robert Frost pointed out that the goal of all reservation or creation is to make benefit. The U.S financial crisis which started in 2007 has proved this statement. To make big benefit, the U.S government, finance institute, mortgage organization and real estate operators have deducted a worldwide risk game in the beginning of this century. The breadth of its impact and the huge damage it brings to global economy are rarely seen in the history of world economy. In the current financial crisis, a large number of corporations get bankrupted. These painful lessons undoubtedly point out one thing: the capability of corporation treasury management has to be pushed up to a higher level. Hence, as the world economy is getting better, it is highly necessary to probe into the enterprises' treasury management strategy in post financial crisis era.

Modeling of a Thermal Spraying Process by Gaussian Chain Graphs

Abstract: The class of chain graph models is considered for the analysis of a data set arising from a thermal spraying process. Different methods for the selection of Gaussian chain graph models are reviewed and applied. Freely available software for the semi-automatic model selection of Gaussian chain graph models is used and compared.

Integer Programming for Bayesian Network Structure Learning

Abstract: Bayesian networks provide an attractive representation of structured probabilistic information. There is thus much interest in 'learning' BNs from data. In this paper the problem of learning a Bayesian network using integer programming is presented. The SCIP (Solving Constraint Integer Programming) framework is used to do this. Although cutting planes are a key ingredient in our approach, primal heuristics and efficient propagation are also important.

High Performance Computing for a Financial Application Using Fast Fourier Transform

Abstract: Fast Fourier Transform (FFT) has been used in many scientific and engineering applications. In the current study, we have applied the FFT for a novel application in finance. We have improved a mathematical model of Fourier transform technique for pricing financial derivatives to help design an effective parallel algorithm. We have then developed a new parallel algorithm for FFT using a swapping technique that exploits data locality. We have analyzed our algorithm theoretically and have reported the significance of the new algorithm. We have implemented our algorithm on 20 node SunFire 6800 high performance computing system and compared the new algorithm with the traditional Cooley-Tukey algorithm both as stand alone comparison of the performance and in relation to our theoretical analysis and showed higher efficiency of our algorithm. We have presented the computed option values for various strike prices with a proper selection of strike-price spacing to ensure fine-grid integration for FFT comput...

Weighted Estimation of Quantiles Using Unbalanced Ranked Set Sampling

Abstract: HeIn the case where the population distribution is unknown, the nonparametric weighted estimation of quantiles under unbalanced ranked set sampling is proposed because observations with different ranks are not identically distributed. It is shown analytically the optimal weighted version always improves the efficiency for all distributions whether ranking is perfect or not. Then, the sampling allocation that maximizes the estimation efficiency is identified and shown to not depend on the population distribution. Finally, results of computed relative efficiencies and an application to a real data set are presented to illustrate some of the theoretical findings.

Objective Bayesian Analysis for Recurrent Events in Presence of Competing Risks

Abstract: HeIn this article, we use Bayesian reference analysis to consider the model of recurrent events with competing risks, where the sampling distribution of the observations due to some failure types i...

Statistical Inference of Weighted Exponential Lifetimes Under Progressive Type-II Censoring Scheme

Abstract: This paper considers estimation of parameters of weighted exponential (WE) distribution based on the progressively Type-II censored data. First the parameters are estimated by the maximum likelihood (MLE) method. It is observed that the MLE of parameters cannot be obtained in a closed form. So, the approximate maximum likelihood estimates (AMLE) approach is proposed to deal with non-linear expressions resulted from the MLE method. A further point estimation method, Bayes estimation, is utilized which does not result in explicit form for the obtained integrals. We use Lindley’s approximation method to get rid of unsolvable integrals designed for squared error and linex loss functions. Also, the Fisher information matrix is found and used to construct asymptotic confidence interval. The two alternative approximate confidence intervals such as percentile bootstrap and bootstrap-t are also derived. Finally, a simulation study in order to compare the proposed estimators is performed.

Setup Adjustment for Asymmetric Cost Functions Under Unknown Process Parameters

Abstract: This paper presents a Bayesian approach for the optimal control of a machine that can experience setup errors assuming an asymmetric off-target cost function. It is assumed that the setup error cannot be observed directly due to presence of measurement and part-to-part errors, and it is further assumed that the variance of this error is not known a priori. The setup error can be compensated by performing sequential adjustments of the process mean based on observations of the parts produced. It is shown how the proposed method converges to the optimal (known variance) trajectory, recovering from a possibly biased initial variance estimate. Simulations results are presented to show the small sample behavior of the proposed method under two types of asymmetric off-target cost functions: a constant asymmetric and a quadratic asymmetric cost function.

Bounded Loss Functions and the Characteristic Function Inversion Method for Computing Expected Loss

Abstract: In robust parameter design, the quadratic loss function is commonly used. However, this loss function is not always realistic and the expected loss may not exist in some cases. This paper proposes the use of a general class of bounded loss functions that are cumulative distribution functions and probability density functions. New loss functions are investigated and the loss functions are shown to yield optimal settings different from those obtained with the quadratic loss. For the class of models that are linear in the noise factors, we give a numerical method based on characteristic function inversion for computing expected loss. The method is quick and accurate; thus, it eases computation of the expected loss and comparison of alternative control factor settings. This method is applicable as long as the distributions chosen to represent the loss function and noise factors have tractable characteristic functions.

Optimal Trivariate Replacement Policies for a Deteriorating System

Abstract: HeIn this paper, three replacement policies for a deteriorating system are studied. Assume that the system after repair is not as good as new. This paper proposes two basic replacement policies and one co-policy in extending the general bivariate replacement policy for such a deteriorating system. The objective of our study is to determine the optimal replacement policies such that the expected cost rates are minimized. The explicit expressions of the expected cost rates are derived, and the corresponding optimal policies can be studied and determined numerically. Our model is a generalization of several traditional models in bivariate replacement literature. Finally, a numerical example is presented for illustrating the theoretical results.

Unified and Dynamic Monitoring of Solvency of Property and Liability Insurance Companies

Abstract: Dynamic monitoring is an accepted and widely used technique of industrial quality control. In this work, we apply dynamic monitoring to monitor and predict insurer financial strength. We propose a new statistic that combines means, variances, and co-variances of multivariate financial indices as a type of quality control tool. We employ data for US property and casualty insurers for the period 2001 through 2010 to determine the control regions, and we provide two examples to illustrate the application of our proposed methodology. We also demonstrate that the proposed methodology is especially suitable to emerging insurance markets, although it is applicable for developed insurance markets as well.

A Mixed-Effect Model for Analyzing Experiments with Multistage Processes

Abstract: In industrial practice, most products are produced by processes that involve multiple stages. In studying multistage processes via designed experiments, some practitioners treat them as single-stage processes and follow the usual factorial designs or split-plot designs. In this paper, through an analysis of the error transmission mechanism, we propose a mixed-effect model for analyzing experiments with multistage processes. Based on an analysis of simulated and real experimental data, we find that different conclusions about factor significance may be drawn if the data are analyzed differently. In addition, the mixed-effect model can help separate errors at different stages and hence provide more information on process improvement.

Generalized Covariation Function for Random Variables with Finite First Moments

Abstract: In this paper, we consider a linear dependence measure between two random variables with finite first moments called as the generalized covariation function. The latter includes the covariation and covariance functions as special cases. We investigate some basic properties of the function and define its moment type estimator. We also investigate the numerical properties of this estimator using the simulated SαS data. We show importance of results in applications, we apply the generalized covariation function to estimate the coefficient “beta” of “stable” CAPM (Belkacem et al. [1]) using real data from Indonesian stock exchange (IDX).

Portfolio Selection with a Hidden Markov Model

Abstract: We consider the problem of investment and consumption with a hidden Markov model and a regime switching structure. The Bayesian approach is followed to integrate econometric consideration and to make inference of the hidden Markov model. The optimal investment strategy is characterized by the method of stochastic dynamic programming and simulation results are given.

Multi-Factor Statistical Modelling of Demand and Spot Price of Electricity

Abstract: The deregulation of electricity markets in different parts of the world has exposed consumers to irregularities in electricity prices driven by the principle of supply and demand. Development of accurate statistical models contributes to establishing protective mechanisms and risk measurement policies for both suppliers, consumers. In this paper multi-factor modelling methodology, solely applied to the spot price of electricity or demand for electricity in earlier studies, is extended to a bivariate process of spot price of electricity and demand for electricity. The suggested model accommodates common idiosyncrasies observed in deregulated electricity markets such as cyclical trends in price and demand for electricity, occurrence of extreme spikes in prices, and a mean-reversion effect seen in the settling of prices from extreme values to the mean level over a short period of time. A time series model for de-seasonalised demand for electricity is used in combination with a linear regression model...

Using Pairwise Ordering Preferences to Estimate Causal Effects in Gene Expression from a Mixture of Observational and Intervention Experiments

Abstract: An important step in systems biology is to improve our knowledge of how genes causally interact with one another. A few approaches have been proposed for the estimation of causal effects among genes, either based on observational data alone or requiring a very precise intervention design with one knock-out experiment for each gene. We recently suggested a more flexible algorithm, using a Markov chain Monte Carlo algorithm and the Mallows ranking model, that can analyze any intervention design, including partial or multiple knock-outs, using the framework of Gaussian Bayesian networks. We previously demonstrated the superior performance of this algorithm in comparison to alternative methods, although it can be computationally expensive to implement. The aim of this paper is to propose an alternative approach taking advantage of node pair ordering preferences to sample the posterior distribution according to the Babington-Smith ranking distribution. This novel algorithm proved, both in a simulation study and on the DREAM4 challenge data, to attain estimation of the causal effects as accurate as the MCMC-Mallows approach with a highly improved computational efficiency, being at least 100 times faster. We also tested our algorithm on the Rosetta Compendium dataset with more contrasted results. We nevertheless anticipate that our new approach might be very useful for practical biological applications.