• Communications for Statistical Applications and Methods
• /
• 제9권3호
• /
• pp.799-808
• /
• 2002

This paper aims at selecting the multi-response design with γ responses minimizing the bias error caused by fitting inadequate models to responses, where the first order models are fitted to Ρ responses fearing the quadratic bias, while to other γ- Ρ responses, the quadratic models are fitted fearing the cubic biases in the cuboidal region of interest. Under the assumption of symmetric design, by minimizing the criterion which represents the amount of error caused by fitting inadequate models, the optimum design was found to be the one having the design moments of second order and the fourth order as 1/3 and l/5, respectively. Examples of the design meeting the required conditions are given for illustration.

• Communications for Statistical Applications and Methods
• /
• 제13권3호
• /
• pp.755-764
• /
• 2006

Most graphical methods for categorical data can describe the structure of data and represent a measure of association among categorical variables. Among them the polyhedron plot represents sequential relationships among hierarchical log-linear models for a multidimensional contingency table. This kind of plot could be explored to describe the differences among sequential models. In this paper we suggest graphical methods, containing all the information, that reflect the relationship among all log-linear models in a certain hierarchical structure. We use the ideas of a correlation diagram.

• Communications for Statistical Applications and Methods
• /
• 제5권3호
• /
• pp.847-853
• /
• 1998

Suppose we want to compare following non-hierarchical log-linear models, $H_0:f(x, heta inTheta_a)$ vs H_1:g(x, heta inTheta_eta); for; Theta_a,;Theta_etasubsetTheta;such;that;Theta_$\alpha$/ Theta_eta$. The goodness of fit test using the likelihood ratio test statistic for comparing these models could not be acceptable. By using the polyhedrons plots of Choi and Hong (1995), we propose a method to decide a better model between two non-hierarchical log-linear models $f(x: heta inTheta_a) and g(x: heta inTheta_eta)$.

• Communications for Statistical Applications and Methods
• /
• 제11권2호
• /
• pp.381-397
• /
• 2004

The marginal likelihood has become an important tool for model selection in Bayesian analysis because it can be used to rank the models. We discuss the marginal likelihood for Poisson regression models that are potentially useful in small area estimation. Computation in these models is intensive and it requires an implementation of Markov chain Monte Carlo (MCMC) methods. Using importance sampling and multivariate density estimation, we demonstrate a computation of the marginal likelihood through an output analysis from an MCMC sampler.

• 한국경영과학회:학술대회논문집
• /
• 한국경영과학회 1996년도 추계학술대회발표논문집; 고려대학교, 서울; 26 Oct. 1996
• /
• pp.39-42
• /
• 1996

Modeling techniques such as linear regression have been used to predict hurricane activity many months in advance of the start of the hurricane season with some success. In this paper, we construct feedforward neural networks to model Atlantic basin hurricane activity and compare the predictions of our neural network models to the predictions produced by statistical models found in the weather forecasting literature. We find that our neural network models produce reasonably accurate predictions that, for the most part, compare favorably to the predictions of statistical models.

• 한국기계가공학회지
• /
• 제7권4호
• /
• pp.128-134
• /
• 2008

Models of the deposition heights in the uncontrolled laser aided direct metal deposition process are constructed for the enhancement of the process integrity. Linear and non-linear statistical models as well as fuzzy model are utilized as the modeling methods. The predictability of the models are evaluated with the values of the sum of square error. The algorithm to use the models in the feedback controlled system is suggested to increase the deposition height accuracy within a layer.

• 한국경영과학회지
• /
• 제18권2호
• /
• pp.57-81
• /
• 1993

This paper is concerned with analyzing the bankruptcy prediction power of three methods : Multivariate Discriminant Analysis (MDA), Inductive Learning, Neural Network, MDA has been famous for its effectiveness for predicting bankrupcy in accounting fields. However, it requires rigorous statistical assumptions, so that violating one of the assumptions may result in biased outputs. In this respect, we alternatively propose the use of two AI models for bankrupcy prediction-inductive learning and neural network. To compare the performance of those two AI models with that of MDA, we have performed massive experiments with a number of Korean bankrupt-cases. Experimental results show that AI models proposed in this study can yield more robust and generalizing bankrupcy prediction than the conventional MDA can do.

• Communications for Statistical Applications and Methods
• /
• 제19권6호
• /
• pp.885-898
• /
• 2012

In this paper, we consider Bayesian approaches to partially linear models, in which a regression function is represented by a semiparametric additive form of a parametric linear regression function and a nonparametric regression function. We make a comparative study on the performance of widely used Bayesian partially linear models in terms of empirical analysis. Specifically, we deal with three Bayesian methods to estimate the nonparametric regression function, one method using Fourier series representation, the other method based on Gaussian process regression approach, and the third method based on the smoothness of the function and differencing. We compare the numerical performance of three methods by the root mean squared error(RMSE). For empirical analysis, we consider synthetic data with simulation studies and real data application by fitting each of them with three Bayesian methods and comparing the RMSEs.

• Communications for Statistical Applications and Methods
• /
• 제21권2호
• /
• pp.161-168
• /
• 2014

In mixture experiments with process variables, we consider the case that some of process variables are either uncontrollable or hard to control, which are called noise variables. Given the such mixture experimental data with process variables, first we study how to search for candidate models. Good candidate models are screened by the sequential variables selection method and checking the residual plots for the validity of the model assumption. Two methods, which use numerical optimization methods proposed by Derringer and Suich (1980) and minimization of the weighted expected loss, are proposed to find a cost-effective robust optimal condition in which the performance of the mean as well as the variance of the response for each of the candidate models is well-behaved under the cost restriction of the mixture. The proposed methods are illustrated with the well known fish patties texture example described by Cornell (2002).

• Communications for Statistical Applications and Methods
• /
• 제10권1호
• /
• pp.239-246
• /
• 2003

Statisticians, or data miners, are often requested to assess the importances of input variables in the given supervised learning model. For the purpose, one may rely on separate ad hoc measures depending on modeling types, such as linear regressions, the neural networks or trees. Consequently, the conceptual consistency in input variable importance measures is lacking, so that the measures cannot be directly used in comparing different types of models, which is often done in data mining processes, In this short communication, we propose a unified approach to the importance measurement of input variables. Our method uses sensitivity analysis which begins by perturbing the values of input variables and monitors the output change. Research scope is limited to the models for continuous output, although it is not difficult to extend the method to supervised learning models for categorical outcomes.