• Journal of Korean Society of Water and Wastewater
• /
• v.35 no.4
• /
• pp.259-275
• /
• 2021

This study explored the usefulness and implications of the Bayesian hyperparameter optimization in developing species distribution models (SDMs). A variety of machine learning (ML) algorithms, namely, support vector machine (SVM), random forest (RF), boosted regression tree (BRT), XGBoost (XGB), and Multilayer perceptron (MLP) were used for predicting the occurrence of four benthic macroinvertebrate species. The Bayesian optimization method successfully tuned model hyperparameters, with all ML models resulting an area under the curve (AUC) > 0.7. Also, hyperparameter search ranges that generally clustered around the optimal values suggest the efficiency of the Bayesian optimization in finding optimal sets of hyperparameters. Tree based ensemble algorithms (BRT, RF, and XGB) tended to show higher performances than SVM and MLP. Important hyperparameters and optimal values differed by species and ML model, indicating the necessity of hyperparameter tuning for improving individual model performances. The optimization results demonstrate that for all macroinvertebrate species SVM and RF required fewer numbers of trials until obtaining optimal hyperparameter sets, leading to reduced computational cost compared to other ML algorithms. The results of this study suggest that the Bayesian optimization is an efficient method for hyperparameter optimization of machine learning algorithms.

• Journal of Environmental Science International
• /
• v.28 no.1
• /
• pp.19-35
• /
• 2019

This study investigates the performance of four Bayesian methods, Random Walk Metropolis (RWM), Hit-And-Run Metropolis (HARM), Adaptive Mixture Metropolis (AMM), and Population Monte Carlo (PMC), for estimating the parameters and uncertainties of probability rainfall distribution, and the results are compared with those of conventional parameter estimation methods; namely, the Method Of Moment (MOM), Maximum Likelihood Method (MLM), and Probability Weighted Method (PWM). As a result, Bayesian methods yield similar or slightly better results in parameter estimations compared with conventional methods. In particular, PMC can reduce parameter uncertainty greatly compared with RWM, HARM, and AMM methods although the Bayesian methods produce similar results in parameter estimations. Overall, the Bayesian methods produce better accuracy for scale parameters compared with the conventional methods and this characteristic improves the accuracy of probability rainfall. Therefore, Bayesian methods can be effective tools for estimating the parameters and uncertainties of probability rainfall distribution in hydrological practices, flood risk assessment, and decision-making support.

• Journal of Korea Water Resources Association
• /
• v.45 no.5
• /
• pp.505-516
• /
• 2012

The study applies a hydrologic simulation model, HEC-1 developed by Hydrologic Engineering Center to Daecheong dam watershed for modeling hourly inflows of Daecheong dam. Although the HEC-1 model provides an automatic optimization technique for some of the parameters, the built-in optimization model is not sufficient in estimating reliable parameters. In particular, the optimization model often fails to estimate the parameters when a large number of parameters exist. In this regard, a main objective of this study is to develop Bayesian Markov Chain Monte Carlo simulation based HEC-1 model (BHEC-1). The Clark IUH method for transformation of precipitation excess to runoff and the soil conservation service runoff curve method for abstractions were used in Bayesian Monte Carlo simulation. Simulations of runoff at the Daecheong station in the HEC-1 model under Bayesian optimization scheme allow the posterior probability distributions of the hydrograph thus providing uncertainties in rainfall-runoff process. The proposed model showed a powerful performance in terms of estimating model parameters and deriving full uncertainties so that the model can be applied to various hydrologic problems such as frequency curve derivation, dam risk analysis and climate change study.

• Journal of KIISE:Software and Applications
• /
• v.29 no.11
• /
• pp.775-784
• /
• 2002

Microarray data, obtained from DNA chip technologies, is the measurement of the expression level of thousands of genes in cells or tissues. It is used for gene function prediction or cancer diagnosis based on gene expression patterns. Among diverse methods for data analysis, the Bayesian network represents the relationships among data attributes in the form of a graph structure. This property enables us to discover various relations among genes and the characteristics of the tissue (e.g., the cancer type) through microarray data analysis. However, most of the present microarray data sets are so sparse that it is difficult to apply general analysis methods, including Bayesian networks, directly. In this paper, we harness an efficient structural learning algorithm and data dimensionality reduction in order to analyze microarray data using Bayesian networks. The proposed method was applied to the analysis of real microarray data, i.e., the NC160 data set. And its usefulness was evaluated based on the accuracy of the teamed Bayesian networks on representing the known biological facts.

• Journal of KIISE:Software and Applications
• /
• v.36 no.3
• /
• pp.208-216
• /
• 2009

A number of estimation of distribution algorithms have been proposed that do not use explicitly crossover and mutation of traditional genetic algorithms, but estimate the distribution of population for more efficient search. But because it is not easy to discover higher-order correlations of variables, lower-order correlations are estimated most cases under various constraints. In this paper, we propose a new estimation of distribution algorithm that represents higher-order correlations of the data and finds global optimum more efficiently. The proposed algorithm represents the higher-order correlations among variables by building random hypergraph model composed of hyperedges consisting of variables which are expected to be correlated, and generates the next population by Bayesian sampling algorithm Experimental results show that the proposed algorithm can find global optimum and outperforms the simple genetic algorithm and BOA(Bayesian Optimization Algorithm) on decomposable functions with deceptive building blocks.

• Korean Journal of Logic
• /
• v.8 no.2
• /
• pp.33-58
• /
• 2005

The old evidence problem raises a profound problem to Bayesian theory of confirmation that evidence known prior to a hypothesis explaining it cannot give any empirical support to the hypothesis. The old evidence problem has resisted to a lot of trials to solve it. The purpose of the paper is to solve the old evidence problem by showing that the problem originated from a serious misunderstanding about the Bayesian concept of confirmation. First, I shall make a brief analysis of the problem, and examine critically two typical Bayesian strategies to solve it. Second, I shah point out a misunderstanding commonly found among Bayesian discussions about the old evidence problem, the ignorance of the asymmetry of confirmation in the context of explanation and prediction. Lastly, 1 shall suggest two different concepts of confirmations by using the asymmetry and argue that the concept of confirmation presupposed in the old evidence problem is not a genuine Bayesian concept of confirmation.

• Journal of Radiation Protection and Research
• /
• v.41 no.2
• /
• pp.149-154
• /
• 2016

Background: Any real application of Bayesian inference must acknowledge that both prior distribution and likelihood function have only been specified as more or less convenient approximations to whatever the analyzer's true belief might be. If the inferences from the Bayesian analysis are to be trusted, it is important to determine that they are robust to such variations of prior and likelihood as might also be consistent with the analyzer's stated beliefs. Materials and Methods: The robust Bayesian inference was applied to atmospheric dispersion assessment using Gaussian plume model. The scopes of contaminations were specified as the uncertainties of distribution type and parametric variability. The probabilistic distribution of model parameters was assumed to be contaminated as the symmetric unimodal and unimodal distributions. The distribution of the sector-averaged relative concentrations was then calculated by applying the contaminated priors to the model parameters. Results and Discussion: The sector-averaged concentrations for stability class were compared by applying the symmetric unimodal and unimodal priors, respectively, as the contaminated one based on the class of ${\varepsilon}$-contamination. Though ${\varepsilon}$ was assumed as 10%, the medians reflecting the symmetric unimodal priors were nearly approximated within 10% compared with ones reflecting the plausible ones. However, the medians reflecting the unimodal priors were approximated within 20% for a few downwind distances compared with ones reflecting the plausible ones. Conclusion: The robustness has been answered by estimating how the results of the Bayesian inferences are robust to reasonable variations of the plausible priors. From these robust inferences, it is reasonable to apply the symmetric unimodal priors for analyzing the robustness of the Bayesian inferences.

• The Korean Journal of Applied Statistics
• /
• v.27 no.6
• /
• pp.1049-1068
• /
• 2014

Autoregressive models are used to analyze an univariate time series data; however, these methods can be inappropriate when a structural break appears in a time series since they assume that a trend is consistent. Threshold autoregressive models (popular regime-switching models) have been proposed to address this problem. Recently, the models have been extended to two regime-switching models with delay parameter. We discuss two regime-switching threshold autoregressive models from a Bayesian point of view. For a Bayesian analysis, we consider a parametric threshold autoregressive model and a nonparametric threshold autoregressive model using Dirichlet process prior. The posterior distributions are derived and the posterior inferences is performed via Markov chain Monte Carlo method and based on two Bayesian threshold autoregressive models. We present a simulation study to compare the performance of the models. We also apply models to gross domestic product data of U.S.A and South Korea.

• The Journal of the Korea Contents Association
• /
• v.7 no.11
• /
• pp.23-33
• /
• 2007

The collaborative filtering has used the nearest neighborhood method based on the preference and the similarity using the Pearson correlation coefficient. Therefore, it does not reflect content of the items and has the problems of the sparsity and scalability as well. the item-based collaborative filtering has been practically used to improve these defects, but it still does not reflect attributes of the item. In this paper, we propose the user and item based collaborative filtering using the classification property and Naive Bayesian to supplement the defects in the existing recommendation system. The proposed method complexity refers to the item similarity based on explicit data and the user similarity based on implicit data for handing the sparse problem. It applies to the Naive Bayesian to the result of reference. Also, it can enhance the accuracy as computation of the item similarity reflects on the correlative rank among the classification property to reflect attributes.

• Journal of Applied Reliability
• /
• v.17 no.3
• /
• pp.213-223
• /
• 2017

Purpose: The purpose of this study is to arrange the life data analysis literatures based on the Bayesian method quantitatively and provide it as tables. Methods: The Bayesian method produces a more accurate estimates of other traditional methods in a small sample size, and it requires specific algorithm and prior information. Based on these three characteristics of the Bayesian method, the criteria for classifying the literature were taken into account. Results: In many studies, there are comparisons of estimation methods for the Bayesian method and maximum likelihood estimation (MLE), and sample size was greater than 10 and not more than 25. In probability distributions, a variety of distributions were found in addition to the distributions of Weibull commonly used in life data analysis, and MCMC and Lindley's Approximation were used evenly. Finally, Gamma, Uniform, Jeffrey and extension of Jeffrey distributions were evenly used as prior information. Conclusion: To verify the characteristics of the Bayesian method which are more superior to other methods in a smaller sample size, studies in less than 10 samples should be carried out. Also, comparative study is required by various distributions, thereby providing guidelines necessary.