• Journal of the Korean Data and Information Science Society
• /
• v.22 no.3
• /
• pp.459-466
• /
• 2011

This article addresses robust Bayesian modeling for meta analysis which derives general conclusion by combining independently performed individual studies. Specifically, we propose hierarchical Bayesian models with unknown variances for meta analysis under priors which are scale mixtures of normal, and thus have tail heavier than that of the normal. For the numerical analysis, we use the Gibbs sampler for calculating Bayesian estimators and illustrate the proposed methods using actual data.

• The Journal of the Acoustical Society of Korea
• /
• v.11 no.6
• /
• pp.46-52
• /
• 1992

본 논문에서는 HMM 의 이산분포를 연속분포로 근사시키는 준 연속분포 HMM 에 의한 한국어 단어인식에 관하여 연구하였다. 이 모델의 생성과정에서는 입력벡터의 출력확률을 혼합 다차원 정규분 포로 가정하여 입력벡터의 확률함수와 코드위드의 심볼출력을 선형결합하므로써, 연속분포 모델로 근사 시켰으며, 단어인식과정에서는 생성모델에 의해 이산분포 모델에서 발생되는 양자와 왜곡을 감소시키므 로써 인식률을 향상시켰다. 이 방법을 평가하기 위하여 DDD 지역명을 대상으로 이산분포 HMM과 준연 속분포 HMM 의 비교실험을 수행하였다. 그 결과 준연속분포 HMM 에 의하여 이산분포 HMM 보다 향상된 인식률을 얻을 수 있었다.

• Communications for Statistical Applications and Methods
• /
• v.18 no.3
• /
• pp.343-355
• /
• 2011

For a data with the assumption of the mixture distribution, it is important to find an appropriate threshold and evaluate its performance. The relationship is found of well-known nine classification accuracy measures such as MVD, Youden's index, the closest-to-(0, 1) criterion, the amended closest-to-(0, 1) criterion, SSS, symmetry point, accuracy area, TA, TR. Then some conditions of these measures are categorized into seven groups. Under the normal mixture assumption, we calculate thresholds based on these measures and obtain the corresponding type I and II errors. We could explore that which classification measure has minimum type I and II errors for estimated mixture distribution to understand the strength and weakness of these classification measures.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2008.05a
• /
• pp.790-794
• /
• 2008

본 연구에서는 국내 분단위 강우자료(MMR)를 이용하여 시간해상도에 따른 강우의 공간상관구조 특성을 검토하였다. 이러한 특성을 파악하기 위해 이변량 혼합분포를 이용하여 강우를 모형화한 후 정규분포와 대수 정규분포를 고려하여 시간해상도별로 공간상관함수를 유도하고 그 변동특성을 파악하였다. 또한 분단위 강우 자료를 호우 발생 특성별(태풍, 장마, 대류성 강우)로 분류하여 이에 대한 공간상관함수를 각각 유도하였다. 이때 시간해상도를 고려하기 위한 대상 집성시간은 1, 2, 3, 5, 10, 30, 60분이고, 대상지점은 중부지역의 27개 우량관측소 지점을 이용하였다. 그 적용 결과 분단위 강우자료의 경우 무강우 자료의 영향이 상대적으로 매우 크게 나타나는 것을 확인할 수 있었다. 공간상관거리는 적용 분포형, 호우 발생 특성에 따라 차이가 있지만 1분의 경우 약 $9{\sim}15km$, 60분의 경우 약 $21{\sim}53km$인 것으로 파악되었다. 또한 강우의 집성시간이 길어질수록 공간상관특성이 상대적으로 뚜렷하게 나타나고 공간상관거리가 길어짐을 확인하였다. 본 연구의 결과는 분단위 강우자료의 관측소 밀도가 시단위 강우자료 관측소에 비해 상대적으로 매우 적음을 나타내며, 분단위 강우자료를 이용하여 지점빈도해석과 같은 공간적인 특성을 분석할 경우 적절한 개선방안이 제시되어야함을 의미하는 것이기도 하다.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.28 no.3
• /
• pp.171-176
• /
• 2016

The distribution shapes of air and water temperatures are basic and essential information, which determine the frequency patterns of their occurrence. It is also very useful to understand the changes in long-term air and water temperatures with respect to climate change. The typical distribution shapes of air and water temperatures cannot be well fitted using widely used/accepted normal distributions because their shapes show multimodal distributions. In this study, Gaussian mixture distributions and kernel distributions are suggested as the more suitable models to fit their distribution shapes. Based on the results, the tail shape exhibits different patterns. The tail is long in higher temperature regions of water temperature distribution and in lower temperature regions of air temperature distribution. These types of shape comparisons can be useful to identify the patterns of long-term air and water temperature changes and the relationship between air and water temperatures. It is nearly impossible to identify change patterns using only mean-temperatures and normal distributions.

• Journal of the Korean Data and Information Science Society
• /
• v.21 no.5
• /
• pp.863-871
• /
• 2010

Until now, many studies related circular data are carried out, but the focuses are mainly on mildly peaked symmetric or asymmetric cases. In this paper we studied a modeling process for sharply peaked asymmetric circular data. By using wrapped Laplace, which was firstly introduced by Jammalamadaka and Kozbowski (2003), and its mixture distributions, we considered the model fitting problem of multi-modal circular data as well as unimodal one. In particular we suggested EM algorithm to find ML estimates of the mixture of wrapped Laplace distributions. Simulation results showed that the suggested EM algorithm is very accurate and useful.

• Journal of Korea Water Resources Association
• /
• v.40 no.12
• /
• pp.943-956
• /
• 2007

Among various statistics, the spatial correlation function, that is "correlogram", is frequently used to evaluate or design the rain gauge network and to model the rainfall field. The spatial correlation structure of rainfall has the significant variation due to many factors. Thus, the variation of spatial correlation structure of rainfall causes serious problems when deciding the spatial correlation function of rainfall within the basin. In this study, the spatial rainfall structure was modeled using bivariate mixed distributions to derive monthly spatial correlograms, based on Gaussian and lognormal distributions. This study derived the correlograms using hourly data of 28 rain gauge stations in the Keum river basin. From the results, we concluded as following; (1) Among three cases (Case A, Case B, Case C) considered, the Case A(+,+) seems to be the most relevant as it is not distorted much by zero measurements. (2) The spatial correlograms based on the lognormal distribution, which is theoretically as well as practically adequate, is better than that based on the Gaussian distribution. (3) The spatial correlation in July exponentially decrease more obviously than those in other months. (4) The spatial correlograms should be derived considering the temporal resolution(hourly, daily, etc) of interest.

• The Korean Journal of Applied Statistics
• /
• v.24 no.3
• /
• pp.547-557
• /
• 2011

In this paper we developed a statistical model for traffic volume data which collected from a spot of specific local state road. One peculiar property of daily traffic data is that it has bimodal shape which have two peaks on times of both going to office and coming back to home. So, various mixture models of circular distribution are suggested for bimodal traffic data and EM algorithms are applied to estimate the parameters of the suggested models. To compare the accuracy of the suggested models, classical regressions with dummy variables are also considered. The suggested models for traffic volumn data can be effectively used to estimate missing values due to measuring instrument disorder.

• Journal of Korea Water Resources Association
• /
• v.52 no.12
• /
• pp.1001-1009
• /
• 2019

Due to recent increasing heavy snow in Korea, the damage caused by heavy snow is also increasing. In Korea, there are many efforts including establishing disaster prevention measures to reduce the damage throughout the country, but it is difficult to establish the design criteria due to the characteristics of heavy snow. In this study, snowfall frequency analysis was performed to estimate design snow depths using observed snow depth data at Jinju, Changwon and Hapcheon stations. The conventional frequency analysis is sometime limted to apply to the snow depth data containing zero values which produce unrealistc estimates of distributon parameters. To overcome this problem, this study employed mixed distributions based on Lognormal, Generalized Pareto (GP), Generalized Extreme Value (GEV), Gamma, Gumbel and Weibull distribution. The results show that the mixed distributions produced smaller design snow depths than single distributions, which indicated that the mixed distributions are applicable and practical to estimate design snow depths.

• Journal of Advanced Marine Engineering and Technology
• /
• v.41 no.1
• /
• pp.111-117
• /
• 2017

The arrival time of rescue ships is very important in the event of distress. This paper presents the development of experimental data to calculate the arrival time of rescue ships. The ship's traffic probability distribution was used. Mokpo Port was selected as the area of study, and AIS data for a 1 year period were used. For the ship's traffic probability distribution, a gateline was established. The lateral range distribution was calculated and fitted to the normal distribution and two Gaussian mixture distributions (GMD2), and each parameter was extracted. After the locations of ${\mu}$, ${\mu}{\pm}1{\sigma}$ of the normal distribution and ${\mu}_1$ of the two Gaussian mixture distribution(GMD2) were set as waypoints, the location and probability were determined. A scenario was established in relation to each type of parameter. Thus, the arrival time can be calculated.