The effect of training data set size and the complexity of the separation function on neural network classification capability: The two-group case

Classification among groups is a crucial problem in managerial decision making. Classification techniques are used in: identifying stressed firms, classifying among consumer types, and rating of firms' bonds, etc. Neural networks are recognized as important and emerging methodologies in the area of classification. In this paper, we study the effect of training sample size and the neural network topology on the classification capability of neural networks. We also compare neural network capabilities with those of commonly used statistical methodologies. Experiments were designed and carried out on two-group classification problems to find answers to these questions. The prediction capability of the neural network models are better than traditional statistical models. The learning capability of the neural networks is improving compared to traditional models because the discriminate function is more complex. For real world classification problems, the usage of neural networks is highly recommended, for two reasons: learning capability and flexibility. Learning capability: Neural network classifies better in sterile experiments as performed in this research. Flexibility: Real life data are rarely not contaminated with noise, such as unknown distributions, and missing variables, etc. Neural networks differ from a statistical model that it is not dependent on any assumption concerning the data set distribution. © 1997 John Wiley & Sons, Inc. Naval Research Logistics 44: 699–717, 1997     

Statistical analysis for masked system life data from Marshall‐Olkin Weibull distribution under progressive hybrid censoring

This paper considers the statistical analysis of masked data in a series system, where the components are assumed to have Marshall‐Olkin Weibull distribution. Based on type‐I progressive hybrid censored and masked data, we derive the maximum likelihood estimates, approximate confidence intervals, and bootstrap confidence intervals of unknown parameters. As the maximum likelihood estimate does not exist for small sample size, Gibbs sampling is used to obtain the Bayesian estimates and Monte Carlo method is employed to construct the credible intervals based on Jefferys prior with partial information. Numerical simulations are performed to compare the performances of the proposed methods and one data set is analyzed.     

An application of empirical bayes techniques to the simultaneous estimation of many probabilities

Consider the following situation: Each of N different combat units is presented with a number of requirements to satisfy, each requirement being classified into one of K mutually exclusive categories. For each unit and each category, an estimate of the probability of that unit satisfying any requirement in that category is desired. The problem can be generally stated as that of estimating N different K-dimensional vectors of probabilities based upon a corresponding set of K-dimensional vectors of sample proportions. An empirical Bayes model is formulated and applied to an example from the Marine Corps Combat Readiness Evaluation System (MCCRES). The EM algorithm provides a convenient method of estimating the prior parameters. The Bayes estimates are compared to the ordinary estimates, i.e., the sample proportions, by means of cross validation, and the Bayes estimates are shown to provide considerable improvement.     

Dichotomous categorical variable formation in mathematical programming discriminant analysis models

Classification models, whether generated by statistical techniques or mathematical programming (MP) discriminant analysis methods, are often simplified by ad hoc formation of dichotomous categorical variables from the original variables with, for example, a dichotomous variable taking value 1 if the original variable is above a threshold level and 0 otherwise. In this paper an MP discriminant analysis method is developed for forming dichotomous categorical variables in problems with discriminant functions that are monotone in the original variables. For each of the original variables from which dichotomous variables may be formed, a set of possible threshold levels for dichotomous variable formation is defined. An MP model is then used to determine both the threshold level for forming each dichotomous variable and the associated discriminant function coefficient. The proposed MP approach is applied to a published problem and a number of simulated problem sets. It is shown that the discriminant functions in dichotomous categorical variables generated by this new MP approach can in some cases outperform the functions generated by standard MP discriminant analysis models using the original variables. © 2004 Wiley Periodicals, Inc. Naval Research Logistics, 2004.     

On the generation of deep disjunctive cutting planes

In this paper we address the question of deriving deep cuts for nonconvex disjunctive programs. These problems include logical constraints which restrict the variables to at least one of a finite number of constraint sets. Based on the works of Balas. Glover, and Jeroslow, we examine the set of valid inequalities or cuts which one may derive in this context, and defining reasonable criteria to measure depth of a cut we demonstrate how one may obtain the “deepest” cut. The analysis covers the case where each constraint set in the logical statement has only one constraint and is also extended for the case where each of these constraint sets may have more than one constraint.     

An integer decomposition algorithm for solving a two‐stage facility location problem with second‐stage activation costs

We study a stochastic scenario‐based facility location problem arising in situations when facilities must first be located, then activated in a particular scenario before they can be used to satisfy scenario demands. Unlike typical facility location problems, fixed charges arise in the initial location of the facilities, and then in the activation of located facilities. The first‐stage variables in our problem are the traditional binary facility‐location variables, whereas the second‐stage variables involve a mix of binary facility‐activation variables and continuous flow variables. Benders decomposition is not applicable for these problems due to the presence of the second‐stage integer activation variables. Instead, we derive cutting planes tailored to the problem under investigation from recourse solution data. These cutting planes are derived by solving a series of specialized shortest path problems based on a modified residual graph from the recourse solution, and are tighter than the general cuts established by Laporte and Louveaux for two‐stage binary programming problems. We demonstrate the computational efficacy of our approach on a variety of randomly generated test problems. © 2010 Wiley Periodicals, Inc. Naval Research Logistics, 2010     

An application of statistical techniques to estimate engineering man-hours on major aircraft programs

Data on 23 lots of various aircraft programs were gathered. Total engineering man-hours, and information on performance, weight, area, avionics systems, data, and schedule were subjected to least squares analysis. An equation is presented which indicates a relationship between total engineering manhours and a set of seven predictor variables. While the equation derived could only be used with confidence by the manufacturer whose data was analyzed, this article should be looked upon as demonstrating a method of data analysis which others may also find useful, not only for predicting engineering manhours in major aircraft programs, but also in other situations where there is an abundance of possible predictor variables, and the problem is to sort out a meaningful subset of these variables. In order to demonstrate the viability of the formula obtained, comparisons were made with various bid programs.     

数字分析法在消防统计中的应用

数据质量不仅是统计工作的生命线,也是进行各种定量分析的重要基础.利用科学的方法诊断统计数据质量已成为统计学界探讨和研究的热门内容之一.Benford定律中第一位数字的分布规律应该满足对数分布,当数据出现质量缺陷或人为修饰后该样本数据将与Benford定律的期望频率出现偏差.将Benford定律引入到消防统计数据进行质量分析,运用Pearson相关系数对《中国消防年鉴》（2012）公布的消防统计数据进行实证分析,结果表明：统计数据质量较高.     

Monitoring poisson count data with probability control limits when sample sizes are time varying

This article considers the problem of monitoring Poisson count data when sample sizes are time varying without assuming a priori knowledge of sample sizes. Traditional control charts, whose control limits are often determined before the control charts are activated, are constructed based on perfect knowledge of sample sizes. In practice, however, future sample sizes are often unknown. Making an inappropriate assumption of the distribution function could lead to unexpected performance of the control charts, for example, excessive false alarms in the early runs of the control charts, which would in turn hurt an operator's confidence in valid alarms. To overcome this problem, we propose the use of probability control limits, which are determined based on the realization of sample sizes online. The conditional probability that the charting statistic exceeds the control limit at present given that there has not been a single alarm before can be guaranteed to meet a specified false alarm rate. Simulation studies show that our proposed control chart is able to deliver satisfactory run length performance for any time‐varying sample sizes. The idea presented in this article can be applied to any effective control charts such as the exponentially weighted moving average or cumulative sum chart. © 2013 Wiley Periodicals, Inc. Naval Research Logistics 60: 625–636, 2013     

Statistical analysis of the output data from terminating simulations

In this paper we precisely define the two types of simulations (terminating and steady-state) with regard to analysis of simulation output and discuss some common measures of performance for each type. In addition, we conclude, on the basis of discussions with many simulation practitioners, that both types of simulations are important in practice. This is contrary to the impression one gets from reading the simulation literature, where the steady-state case is almost exclusively considered. Although analyses of terminating simulations are considerably easier than are those of steady-state simulations, they have not received a careful treatment in the literature. We discuss and give empirical results for fixed sample size, relative width, and absolute width procedures that can be used for constructing confidence intervals for measures of performance in the terminating case.     

Single‐commodity stochastic network design under demand and topological uncertainties with insufficient data

Stochastic network design is fundamental to transportation and logistic problems in practice, yet faces new modeling and computational challenges resulted from heterogeneous sources of uncertainties and their unknown distributions given limited data. In this article, we design arcs in a network to optimize the cost of single‐commodity flows under random demand and arc disruptions. We minimize the network design cost plus cost associated with network performance under uncertainty evaluated by two schemes. The first scheme restricts demand and arc capacities in budgeted uncertainty sets and minimizes the worst‐case cost of supply generation and network flows for any possible realizations. The second scheme generates a finite set of samples from statistical information (e.g., moments) of data and minimizes the expected cost of supplies and flows, for which we bound the worst‐case cost using budgeted uncertainty sets. We develop cutting‐plane algorithms for solving the mixed‐integer nonlinear programming reformulations of the problem under the two schemes. We compare the computational efficacy of different approaches and analyze the results by testing diverse instances of random and real‐world networks. © 2017 Wiley Periodicals, Inc. Naval Research Logistics 64: 154–173, 2017     

On the reliability,availability and bayes confidence intervals for multicomponent systems

The problem of computing reliability and availability and their associated confidence limits for multi-component systems has appeared often in the literature. This problem arises where some or all of the component reliabilities and availabilities are statistical estimates (random variables) from test and other data. The problem of computing confidence limits has generally been considered difficult and treated only on a case-by-case basis. This paper deals with Bayes confidence limits on reliability and availability for a more general class of systems than previously considered including, as special cases, series-parallel and standby systems applications. The posterior distributions obtained are exact in theory and their numerical evaluation is limited only by computing resources, data representation and round-off in calculations. This paper collects and generalizes previous results of the authors and others. The methods presented in this paper apply both to reliability and availability analysis. The conceptual development requires only that system reliability or availability be probabilities defined in terms acceptable for a particular application. The emphasis is on Bayes Analysis and the determination of the posterior distribution functions. Having these, the calculation of point estimates and confidence limits is routine. This paper includes several examples of estimating system reliability and confidence limits based on observed component test data. Also included is an example of the numerical procedure for computing Bayes confidence limits for the reliability of a system consisting of N failure independent components connected in series. Both an exact and a new approximate numerical procedure for computing point and interval estimates of reliability are presented. A comparison is made of the results obtained from the two procedures. It is shown that the approximation is entirely sufficient for most reliability engineering analysis.     

基于粗糙集的空袭目标威胁评估排序

通过对影响空袭目标威胁排序因素的分析,针对定性与定量因素并存的情况,提出利用粗糙集理论进行威胁排序的方法.将已有决策表转化为成对比较表,利用优势属性集来推理决策规则,从而确定各目标的排序.案例表明此方法有效地解决了空袭目标威胁排序中的如何处理不确定信息的问题,大大提高了决策的准确性和客观性.     

Algorithm to solve a chance‐constrained network capacity design problem with stochastic demands and finite support

We consider the problem of determining the capacity to assign to each arc in a given network, subject to uncertainty in the supply and/or demand of each node. This design problem underlies many real‐world applications, such as the design of power transmission and telecommunications networks. We first consider the case where a set of supply/demand scenarios are provided, and we must determine the minimum‐cost set of arc capacities such that a feasible flow exists for each scenario. We briefly review existing theoretical approaches to solving this problem and explore implementation strategies to reduce run times. With this as a foundation, our primary focus is on a chance‐constrained version of the problem in which α% of the scenarios must be feasible under the chosen capacity, where α is a user‐defined parameter and the specific scenarios to be satisfied are not predetermined. We describe an algorithm which utilizes a separation routine for identifying violated cut‐sets which can solve the problem to optimality, and we present computational results. We also present a novel greedy algorithm, our primary contribution, which can be used to solve for a high quality heuristic solution. We present computational analysis to evaluate the performance of our proposed approaches. © 2016 Wiley Periodicals, Inc. Naval Research Logistics 63: 236–246, 2016     

The asymptotic sufficiency of sparse order statistics in tests of fit with nuisance parameters

In an earlier paper, it was shown that for the problem of testing that a sample comes from a completely specified distribution, a relatively small number of order statistics is asymptotically sufficient, and for all asymptotic probability calculations the joint distribution of these order statistics can be assumed to be normal. In the present paper, these results are extended to certain cases where the problem is to test the hypothesis that a sample comes from a distribution which is a member of a specified parametric family of distributions, with the parameters unspecified.     

Parametric analysis of linear programs with upper bounded variables

Linear programming problems with upper bounded variables can be solved by regular simplex method by considering upper bounding constraints as explicit constraints of the problem. However, more efficient methods exist which consider these upper bound constraints implicitly. When parametric analysis for problems with upper bounds is to be carried out, one can use the regular parameter analysis by considering the upper bound constraints explicitly. This paper develops formulas for parametric analysis where upper bound constraints are used implicitly, thus reducing the size of the basic matrix.     

Testing goodness of fit to the increasing failure rate family with censored data

One important thrust in the reliability literature is the development of statistical procedures under various “restricted family” model assumptions such as the increasing failure rate (IFR) and decreasing failure rate (DFR) distributions. However, relatively little work has been done on the problem of testing fit to such families as a null hypothesis. Barlow and Campo proposed graphical methods for assessing goodness of fit to the IFR model in single-sample problems. For the same problem with complete data, Tenga and Santner studied several analytic tests of the null hypothesis that the common underlying distribution is IFR versus the alternative that it is not IFR for complete data. This article considers the same problem for the case of four types of censored data: (i) Type I (time) censoring, (ii) Type I1 (order statistic) censoring, (iii) a hybrid of Type I and Type I1 censoring, and (iv) random censorship. The least favorable distributions of several intuitive test statistics are derived for each of the four types of censoring so that valid small-sample-size α tests can be constructed from them. Properties of these tests are investigated.     

Optimal schedules for special structure flowshops

This paper considers the problem of scheduling a given number of jobs on a specified number of machines in a flowshop where the objective function is to minimize the total throughput time in which all jobs complete processing on all machines. Based on the combinatorial analysis of the problem, several simple algorithms are developed for solving special structure flowshop scheduling problems where the process times are not completely random, but bear a well-defined relationship to one another. The proposed algorithms are both simple and computationally efficient and can optimally solve large-sized problems even with manual computational devices.     

Statistical decision analysis of stochastic linear programming problems

This paper presents a statistical decision analysis of a one-stage linear programming problem with deterministic constraints and stochastic criterion function. Procedures for obtaining numerical results are given which are applicable to any problem having this general form. We begin by stating the statistical decision problems to be considered, and then discuss the expected value of perfect information and the expected value of sample information. In obtaining these quantities, use is made of the distribution of the optimal value of the linear programming problem with stochastic criterion function, and so we discuss Monte Carlo and numerical integration procedures for estimating the mean of this distribution. The case in which the random criterion vector has a multivariate Normal distribution is discussed separately, and more detailed methods are offered. We discuss dual problems, including some relationships of this work with other work in probabilistic linear programming. An example is given in Appendix A showing application of the methods to a sample problem. In Appendix B we consider the accuracy of a procedure for approximating the expected value of information.     

基因表达缺失值的加权回归估计算法

在基因芯片实验中,数据缺失客观存在,并在一定程度上影响芯片数据后续分析结果的准确性。在不增加实验次数的情况下,缺失值估计是降低缺失数据对后续分析影响的有效方法。利用相似性信息的核加权函数来实现缺失值回归估计的局部化,提出了基于加权回归估计的基因表达缺失值估计算法。在两个不同类型的基因芯片数据上,将新方法与几种已知的方法进行了比较分析。实验结果表明,新的估计算法具有比传统缺失值估计算法更好的稳定性和估计准确度。