• Communications for Statistical Applications and Methods
• /
• v.9 no.3
• /
• pp.809-824
• /
• 2002

we propose two methods which separate the variable selection step and the split-point selection step. We call these two algorithms as CHITES method and F&CHITES method. They adapted some of the best characteristics of CART, CHAID, and QUEST. In the first step the variable, which is most significant to predict the target class values, is selected. In the second step, the exhaustive search method is applied to find the splitting point based on the selected variable in the first step. We compared the proposed methods, CART, and QUEST in terms of variable selection bias and power, error rates, and training times. The proposed methods are not only unbiased in the null case, but also powerful for selecting correct variables in non-null cases.

• Communications for Statistical Applications and Methods
• /
• v.7 no.2
• /
• pp.617-631
• /
• 2000

This paper is concerned with suggesting a Bayesian method for variable selection in generalized logit model. It is based on Laplace-Metropolis algorithm intended to propose a simple method for estimating the marginal likelihood of the model. The algorithm then leads to a criterion for the selection of variables. The criterion is to find a subset of variables that maximizes the marginal likelihood of the model and it is seen to be a Bayes rule in a sense that it minimizes the risk of the variable selection under 0-1 loss function. Based upon two examples, the suggested method is illustrated and compared with existing frequentist methods.

• THE INTERNATIONAL COMMERCE & LAW REVIEW
• /
• v.56
• /
• pp.3-27
• /
• 2012

The purpose of this thesis is to develop a relational model which can explain consumer selection for exporting goods and analyze the effect of corporate image on the relations between country image and consumer selection under global trade environment. The specific objectives are as follows: 1) to suggest a concept of consumer selection and measurement criteria, 2) to analyze correlations among country image, corporate image and consumer selection and 3) to find out the effect of corporate image on the relations between country image and consumer selection. The SPSS program for window and LISREL program were used to analyze the data for this study. The statistical method used in this study was the covariance structure analysis estimating parameters by maximum likelihood method. Path coefficients were tested for t-tests with a statistical significance level of .05. The conclusions of this study are as follows. First, significant correlations were observed among all sub-variables proposed in this study. In addition, significant correlations were detected among country image, consumer selection and corporate image. Second, a hypothetical model proposed in this study was mostly appropriate. Country image had a positive direct effect on consumer selection and corporate image with statistical significance. In addition, it has an indirect impact on consumer selection with statistical significance with corporate image as an intervening variable. Third, corporate image had a significant moderation effect in country image-consumer selection relations. As corporate image levels increased, the effect of country image on consumer selection increased as well. In other words, it has been confirmed that if corporate image levels are high, country image could end up with consumer selection.

• Communications for Statistical Applications and Methods
• /
• v.4 no.1
• /
• pp.177-183
• /
• 1997

A data-adaptive order selection procedure is proposed for local polynomial nonparametric regression. For each given polynomial order, bias and variance are estimated and the adaptive polynomial order that has the smallest estimated mean squared error is selected locally at each location point. To estimate mean squared error, empirical bias estimate of Ruppert (1995) and local polynomial variance estimate of Ruppert, Wand, Wand, Holst and Hossjer (1995) are used. Since the proposed method does not require fitting polynomial model of order higher than the model order, it is simpler than the order selection method proposed by Fan and Gijbels (1995b).

• Communications for Statistical Applications and Methods
• /
• v.26 no.2
• /
• pp.149-161
• /
• 2019

In this article, we suggest the following approaches to simultaneous variable selection and outlier detection. First, we determine possible candidates for outliers using properties of an intercept estimator in a difference-based regression model, and the information of outliers is reflected in the multiple regression model adding mean shift parameters. Second, we select the best model from the model including the outlier candidates as predictors using stochastic search variable selection. Finally, we evaluate our method using simulations and real data analysis to yield promising results. In addition, we need to develop our method to make robust estimates. We will also to the nonparametric regression model for simultaneous outlier detection and variable selection.

• Journal of the Korean Statistical Society
• /
• v.22 no.2
• /
• pp.361-368
• /
• 1993

In recent years, kernel type estimates are abundant. In this paper, we propose a bandwidth selection method for kernel regression of fixed design based on bootstrap procedure. Mathematical properties of proposed bootstrap-based bandwidth selection method are discussed. Performance of the proposed method for small sample case is compared with that of cross-validation method via a simulation study.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.10
• /
• pp.3972-3988
• /
• 2020

Existing methods always rely on statistical features to extract local words for microblog user geolocation. There are many non-local words in extracted words, which makes geolocation accuracy lower. Considering the statistical and semantic features of local words, this paper proposes a microblog user geolocation method by extracting local words based on word clustering and wrapper feature selection. First, ordinary words without positional indications are initially filtered based on statistical features. Second, a word clustering algorithm based on word vectors is proposed. The remaining semantically similar words are clustered together based on the distance of word vectors with semantic meanings. Next, a wrapper feature selection algorithm based on sequential backward subset search is proposed. The cluster subset with the best geolocation effect is selected. Words in selected cluster subset are extracted as local words. Finally, the Naive Bayes classifier is trained based on local words to geolocate the microblog user. The proposed method is validated based on two different types of microblog data - Twitter and Weibo. The results show that the proposed method outperforms existing two typical methods based on statistical features in terms of accuracy, precision, recall, and F1-score.

• Journal of the Korean Data and Information Science Society
• /
• v.27 no.6
• /
• pp.1653-1660
• /
• 2016

Smoothly clipped absolute deviation (SCAD) penalty is known to satisfy the desirable properties for penalty functions like as unbiasedness, sparsity and continuity. In this paper, we deal with the regression function estimation and variable selection based on SCAD penalized censored regression model. We use the local linear approximation and the iteratively reweighted least squares algorithm to solve SCAD penalized log likelihood function. The proposed method provides an efficient method for variable selection and regression function estimation. The generalized cross validation function is presented for the model selection. Applications of the proposed method are illustrated through the simulated and a real example.

• Communications for Statistical Applications and Methods
• /
• v.24 no.6
• /
• pp.673-683
• /
• 2017

The least absolute shrinkage and selection operator (LASSO) has been a popular regression estimator with simultaneous variable selection. However, LASSO does not have the oracle property and its robust version is needed in the case of heavy-tailed errors or serious outliers. We propose a robust penalized regression estimator which provide a simultaneous variable selection and estimator. It is based on the rank regression and the non-convex penalty function, the smoothly clipped absolute deviation (SCAD) function which has the oracle property. The proposed method combines the robustness of the rank regression and the oracle property of the SCAD penalty. We develop an efficient algorithm to compute the proposed estimator that includes a SCAD estimate based on the local linear approximation and the tuning parameter of the penalty function. Our estimate can be obtained by the least absolute deviation method. We used an optimal tuning parameter based on the Bayesian information criterion and the cross validation method. Numerical simulation shows that the proposed estimator is robust and effective to analyze contaminated data.

• Communications for Statistical Applications and Methods
• /
• v.10 no.3
• /
• pp.879-894
• /
• 2003

We evaluated the efficiencies of applying attribute selection methods and prior discretization to supervised learning, modelled by C4.5 and Naive Bayes. Three databases were obtained from UCI data archive, which consisted of continuous attributes except for one decision attribute. Four methods were used for attribute selection : MDI, ReliefF, Gain Ratio and Consistency-based method. MDI and ReliefF can be used for both continuous and discrete attributes, but the other two methods can be used only for discrete attributes. Discretization was performed using the Fayyad and Irani method. To investigate the effect of noise included in the database, noises were introduced into the data sets up to the extents of 10 or 20%, and then the data, including those either containing the noises or not, were processed through the steps of attribute selection, discretization and classification. The results of this study indicate that classification of the data based on selected attributes yields higher accuracy than in the case of classifying the full data set, and prior discretization does not lower the accuracy.