• The Korean Journal of Applied Statistics
• /
• v.31 no.2
• /
• pp.265-285
• /
• 2018

In this study, we applied a deep neural network model to predict four grades of fine dust $PM_{10}$, 'Good, Moderate, Bad, Very Bad' and two grades, 'Good or Moderate and Bad or Very Bad'. The deep neural network model and existing classification techniques (such as neural network model, multinomial logistic regression model, support vector machine, and random forest) were applied to fine dust daily data observed from 2010 to 2015 in six major metropolitan areas of Korea. Data analysis shows that the deep neural network model outperforms others in the sense of accuracy.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.05a
• /
• pp.81-84
• /
• 2003

$\varepsilon$-SVR(e-Support Vector Regression)학습방법은 SV(Support Vector)들을 이용하여 함수 근사(Regression)하는 방법으로 최근 주목받고 있는 기법이다. SVM(SV machine)의 한 가지 방법으로, 신경망을 기반으로 한 다른 알고리즘들이 학습과정에서 지역적 최적해로 수렴하는 등의 문제를 한계로 갖는데 반해, 이러한 구조들을 대체할 수 있는 학습방법으로 사용될 수 있다. 일반적인 $\varepsilon$-SVR에서는 학습 데이터와 관사 함수 f사이에 허용 가능한 에러범위 $\varepsilon$값이 학습하기 전에 정해진다. 그러나 Nu-SVR(ν-version SVR)학습방법은 학습의 결과로 최적화 된 $\varepsilon$값을 얻을 수 있다. 정해진 기저함수가 포함되는 $\varepsilon$-SVR 학습방법(Sermparametric SVR)은 정해진 독립 기저함수를 사용하여 함수를 근사하는 방법으로, 일반적인 $\varepsilon$-SVR 학습방범에 비해 우수한 결과를 나타내는 것이 성공적으로 입증된 바 있다. 이에 따라, 본 논문에서는 정해진 기저함수가 포함된 ν-SVR 학습 방법을 제안하고, 이에 대한 수식을 유도하였다. 그리고, 모의 실험을 통하여 제안된 Sermparametric ν-SVR 학습 방법의 적용 가능성을 알아보았다.

• Journal of the Korean Data and Information Science Society
• /
• v.15 no.2
• /
• pp.477-483
• /
• 2004

One of the major paradigms for supervised learning in neural network community is back-propagation learning. The standard implementations of back-propagation learning are optimal under the assumptions of identical and independent Gaussian noise. In this paper, for regression function estimation, we introduce $\varepsilon-insensitive$ back-propagation learning algorithm, which corresponds to minimizing the least absolute error. We compare this algorithm with support vector machine(SVM), which is another $\varepsilon-insensitive$ supervised learning algorithm and has been very successful in pattern recognition and function estimation problems. For comparison, we consider a more realistic model would allow the noise variance itself to depend on the input variables.

• Communications for Statistical Applications and Methods
• /
• v.6 no.3
• /
• pp.813-820
• /
• 1999

Supprot vector machine(SVM) is a new and very promising regression and classification technique developed by Vapnik and his group at AT&T Bell Laboratories. in this paper we provide a brief overview of SVM for regression. Furthermore we describe Bayesian model selection based on macKay's evidence framework for SVM regression.

• Journal of the Korean Data and Information Science Society
• /
• v.23 no.4
• /
• pp.739-748
• /
• 2012

In this paper we propose a mixed effects least squares support vector machine (LS-SVM) for the censored data which are observed from different groups. We use weights by which the randomly right censoring is taken into account in the nonlinear regression. The weights are formed with Kaplan-Meier estimates of censoring distribution. In the proposed model a random effects term representing inter-group variation is included. Furthermore generalized cross validation function is proposed for the selection of the optimal values of hyper-parameters. Experimental results are then presented which indicate the performance of the proposed LS-SVM by comparing with a standard LS-SVM for the censored data.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.250-255
• /
• 2003

The position of a 3-dimensional(3D) point can be measured by using calibrated stereo camera. To obtain more accurate measurement ,more accurate camera calibration is required. There are many existing methods to calibrate camera. The simple linear methods are usually not accurate due to nonlinear lens distortion. The nonlinear methods are accurate more than linear method, but it increase computational cost and good initial guess is needed. The multi step methods need to know some camera parameters of used camera. Recent years, these explicit model based camera calibration work with the development of more precise camera models involving correction of lens distortion. But these explicit model based camera calibration have disadvantages. So implicit camera calibration methods have been derived. One of the popular implicit camera calibration method is to use neural network. In this paper, we propose implicit stereo camera calibration method for 3D reconstruction using support vector machine. SVM can learn the relationship between 3D coordinate and image coordinate, and it shows the robust property with the presence of noise and lens distortion, results of simulation are shown in section 4.

• Geomechanics and Engineering
• /
• v.36 no.4
• /
• pp.381-390
• /
• 2024

The introduction of bio-based materials has been recommended in the geotechnical engineering field to reduce environmental pollutants such as heavy metals and greenhouse gases. However, bio-treated soil methods face limitations in field application due to short research periods and insufficient verification of engineering performance, especially when compared to conventional materials like cement. Therefore, this study aimed to develop a machine learning model for predicting the unconfined compressive strength, a representative soil property, of biopolymer-based soil treatment (BPST). Four machine learning algorithms were compared to determine a suitable model, including linear regression (LR), support vector regression (SVR), random forest (RF), and neural network (NN). Except for LR, the SVR, RF, and NN algorithms exhibited high predictive performance with an R² value of 0.98 or higher. The permutation feature importance technique was used to identify the main factors affecting the strength enhancement of BPST. The results indicated that the unconfined compressive strength of BPST is affected by mean particle size, followed by biopolymer content and water content. With a reliable prediction model, the proposed model can present guidelines prior to laboratory testing and field application, thereby saving a significant amount of time and money.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.45 no.1
• /
• pp.71-82
• /
• 2022

Most of the predictions using machine learning are neutral predictions considering the symmetrical situation where the predicted value is not smaller or larger than the actual value. However, in some situations, asymmetric prediction such as over-prediction or under-prediction may be better than neutral prediction, and it can induce better judgment by providing various predictions to decision makers. A method called Asymmetric Twin Support Vector Regression (ATSVR) using TSVR(Twin Support Vector Regression), which has a fast calculation time, was proposed by controlling the asymmetry of the upper and lower widths of the ε-tube and the asymmetry of the penalty with two parameters. In addition, by applying the existing GSVQR and the proposed ATSVR, prediction using the prediction propensities of over-prediction, under-prediction, and neutral prediction was performed. When two parameters were used for both GSVQR and ATSVR, it was possible to predict according to the prediction propensity, and ATSVR was found to be more than twice as fast in terms of calculation time. On the other hand, in terms of accuracy, there was no significant difference between ATSVR and GSVQR, but it was found that GSVQR reflected the prediction propensity better than ATSVR when checking the figures. The accuracy of under-prediction or over-prediction was lower than that of neutral prediction. It seems that using both parameters rather than using one of the two parameters (p_1,p_2) increases the change in the prediction tendency. However, depending on the situation, it may be better to use only one of the two parameters.

• Computers and Concrete
• /
• v.33 no.6
• /
• pp.739-754
• /
• 2024

This study presents an innovative AI-driven approach to assess the ultimate axial load in Double-Skinned Profiled Steel sheet Composite Walls (DPSCWs). Utilizing a dataset of 80 entries, seven input parameters were employed, and various AI techniques, including Linear Regression, Polynomial Regression, Support Vector Regression, Decision Tree Regression, Decision Tree with AdaBoost Regression, Random Forest Regression, Gradient Boost Regression Tree, Elastic Net Regression, Ridge Regression, and LASSO Regression, were evaluated. Decision Tree Regression and Random Forest Regression emerged as the most accurate models. The top three performing models were integrated into a hybrid approach, excelling in accurately estimating DPSCWs' ultimate axial load. This adaptable hybrid model outperforms traditional methods, reducing errors in complex scenarios. The validated Artificial Neural Network (ANN) model showcases less than 1% error, enhancing reliability. Correlation analysis highlights robust predictions, emphasizing the importance of steel sheet thickness. The study contributes insights for predicting DPSCW strength in civil engineering, suggesting optimization and database expansion. The research advances precise load capacity estimation, empowering engineers to enhance construction safety and explore further machine learning applications in structural engineering.

• Journal of Information Processing Systems
• /
• v.14 no.6
• /
• pp.1385-1397
• /
• 2018

There are many factors that affect the wind speed. In addition, the randomness of wind speed also leads to low prediction accuracy for wind speed. According to this situation, this paper constructs the short-time forecasting model based on the least squares support vector machines (LSSVM) to forecast the wind speed. The basis of the model used in this paper is support vector regression (SVR), which is used to calculate the regression relationships between the historical data and forecasting data of wind speed. In order to improve the forecast precision, historical data is clustered by cluster analysis so that the historical data whose changing trend is similar with the forecasting data can be filtered out. The filtered historical data is used as the training samples for SVR and the parameters would be optimized by particle swarm optimization (PSO). The forecasting model is tested by actual data and the forecast precision is more accurate than the industry standards. The results prove the feasibility and reliability of the model.