• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권11호
• /
• pp.4557-4572
• /
• 2020

The Alaska competition provides an opportunity to study the practical problems of real-world steganalysis. Participants are required to solve steganalysis involving various embedding schemes, inconsistency JPEG Quality Factor and various processing pipelines. In this paper, we propose a method to ensemble multiple deep learning steganalyzers. We select SRNet and RESDET as our base models. Then we design a three-layers model ensemble network to fuse these base models and output the final prediction. By separating the three colors channels for base model training and feature replacement strategy instead of simply merging features, the performance of the model ensemble is greatly improved. The proposed method won second place in the Alaska 1 competition in the end.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2023년도 학술발표회
• /
• pp.25-25
• /
• 2023

The ensemble optimal interpolation (EnOI) scheme is a sub-optimal alternative to the ensemble Kalman filter (EnKF) with a reduced computational demand making it potentially more suitable for operational applications. Since only one model is integrated forward instead of an ensemble of model realizations, online estimation of the background error covariance matrix is not possible in the EnOI scheme. In this study, we investigate two Gaussian noise based ensemble generation strategies to produce dynamic covariance matrices for assimilation of water level observations into a distributed hydrological model. In the first approach, spatially correlated noise, sampled from a normal distribution with a fixed fractional error parameter (which controls its standard deviation), is added to the model forecast state vector to prepare the ensembles. In the second method, we use an adaptive error estimation technique based on the innovation diagnostics to estimate this error parameter within the assimilation framework. The results from a real and a set of synthetic experiments indicate that the EnOI scheme can provide better results when an optimal EnKF is not identified, but performs worse than the ensemble filter when the true error characteristics are known. Furthermore, while the adaptive approach is able to reduce the sensitivity to the fractional error parameter affecting the first (non-adaptive) approach, results are usually worse at ungauged locations with the former.

• Asia-Pacific Journal of Atmospheric Sciences
• /
• 제54권4호
• /
• pp.611-622
• /
• 2018

Observed climate data are processed under the assumption that their time series are stationary, as in multi-step temperature and precipitation prediction, which usually leads to low prediction accuracy. If a climate system model is based on a single prediction model, the prediction results contain significant uncertainty. In order to overcome this drawback, this study uses a method that integrates ensemble prediction and a stepwise regression model based on a mean-valued generation function. In addition, it utilizes empirical mode decomposition (EMD), which is a new method of handling time series. First, a non-stationary time series is decomposed into a series of intrinsic mode functions (IMFs), which are stationary and multi-scale. Then, a different prediction model is constructed for each component of the IMF using numerical ensemble prediction combined with stepwise regression analysis. Finally, the results are fit to a linear regression model, and a short-term climate prediction system is established using the Visual Studio development platform. The model is validated using temperature data from February 1957 to 2005 from 88 weather stations in Guangxi, China. The results show that compared to single-model prediction methods, the EMD and ensemble prediction model is more effective for forecasting climate change and abrupt climate shifts when using historical data for multi-step prediction.

• Journal of the Korean Data and Information Science Society
• /
• 제24권6호
• /
• pp.1113-1125
• /
• 2013

기업의 부도를 예측하는 것은 회계나 재무 분야에서 중요한 연구주제이다. 지금까지 기업 부도예측을 위해 여러 가지 데이터마이닝 기법들이 적용되었으나 주로 단일 모형을 사용함으로서 복잡한 분류 문제에의 적용에 한계를 갖고 있었다. 본 논문에서는 최근에 각광받고 있는 SVM (support vector machine) 모형들을 결합한 앙상블 SVM 모형 (ensemble SVM model)을 부도예측에 사용하고자 한다. 제안된 앙상블 모형은 v-조각 교차 타당성 (v-fold cross-validation)에 의해 얻어진 여러 가지 모형 중에서 성능이 좋은 상위 k개의 단일 모형으로 구성하고 과반수 투표 방식 (majority voting)을 사용하여 미지의 클래스를 분류한다. 본 논문에서 제안된 앙상블 SVM 모형의 성능을 평가하기 위해 실제 기업의 재무비율 자료와 모의실험자료를 가지고 실험하였고, 실험결과 제안된 앙상블 모형이 여러 가지 평가척도 하에서 단일 SVM 모형들보다 좋은 성능을 보임을 알 수 있었다.

• 대기
• /
• 제30권2호
• /
• pp.103-113
• /
• 2020

In this study, we analyzed and developed the monitoring system in order to confirm the effect of observations on forecast sensitivity on ensemble-based data assimilation. For this purpose, we developed the Ensemble Forecast Sensitivity to observation (EFSO) monitoring system based on Local Ensemble Transform Kalman Filter (LETKF) system coupled with Korean Integrated Model (KIM). We calculated 24 h error variance of each of observations and then classified as beneficial or detrimental effects. In details, the relative rankings were according to their magnitude and analyzed the forecast sensitivity by region for north, south hemisphere and tropics. We performed cycle experiment in order to confirm the EFSO result whether reliable or not. According to the evaluation of the EFSO monitoring, GPSRO was classified as detrimental observation during the specified period and reanalyzed by data-denial experiment. Data-denial experiment means that we detect detrimental observation using the EFSO and then repeat the analysis and forecast without using the detrimental observations. The accuracy of forecast in the denial of detrimental GPSRO observation is better than that in the default experiment using all of the GPSRO observation. It means that forecast skill score can be improved by not assimilating observation classified as detrimental one by the EFSO monitoring system.

• 상하수도학회지
• /
• 제28권4호
• /
• pp.377-389
• /
• 2014

In recent years, Smart Water Grid (SWG) concept has globally emerged over the last decade and also gained significant recognition in South Korea. Especially, there has been growing interest in water demand forecast and this has led to various studies regarding energy saving and improvement of water supply reliability. In this regard, this study aims to develop a nonlinear ensemble model for hourly water demand forecasting which allow us to estimate uncertainties across different model classes. The concepts was demonstrated through application to observed from water plant (A) in the South Korea. Various statistics (e.g. the efficiency coefficient, the correlation coefficient, the root mean square error, and a maximum error rate) were evaluated to investigate model efficiency. The ensemble based model with an cross-validate prediction procedure showed better predictability for water demand forecasting at different temporal resolutions. In particular, the performance of the ensemble model on hourly water demand data showed promising results against other individual prediction schemes.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권2호
• /
• pp.632-657
• /
• 2022

Detecting web attacks is a major challenge, and it is observed that the use of simple models leads to low sensitivity or high false positive problems. In this study, we aim to develop a robust two-stage deep learning based stacked ensemble web application firewall. Normal and abnormal classification is carried out in the first stage of the proposed WAF model. The classification process of the types of abnormal traffics is postponed to the second stage and carried out using an integrated stacked ensemble model. By this way, clients' requests can be served without time delay, and attack types can be detected with high sensitivity. In addition to the high accuracy of the proposed model, by using the statistical similarity and diversity analyses in the study, high generalization for the ensemble model is achieved. Within the study, a comprehensive, up-to-date, and robust multi-class web anomaly dataset named GAZI-HTTP is created in accordance with the real-world situations. The performance of the proposed WAF model is compared to state-of-the-art deep learning models and previous studies using the benchmark dataset. The proposed two-stage model achieved multi-class detection rates of 97.43% and 94.77% for GAZI-HTTP and ECML-PKDD, respectively.

• 한국전자거래학회지
• /
• 제25권3호
• /
• pp.1-13
• /
• 2020

기계학습(machine learning)이란 주어진 데이터에 대한 일반화 과정으로부터 특정 문제를 해결할 수 있는 모델(model) 생성 기술을 의미한다. 우수한 성능의 모델을 생성하기 위해서는 양질의 학습데이터와 일반화 과정을 위한 학습 알고리즘이 준비되어야 한다. 성능 개선을 위한 한 가지 방법으로서 앙상블(Ensemble) 기법은 단일 모델(single model)을 생성하기보다 다중 모델을 생성하며, 이는 배깅(Bagging), 부스팅(Boosting), 스태킹(Stacking) 학습 기법을 포함한다. 본 논문은 기존 스태킹 기법을 개선한 다중 스태킹 앙상블(Multiple Stacking Ensemble) 학습 기법을 제안한다. 다중 스태킹 앙상블 기법의 학습 구조는 딥러닝 구조와 유사하고 각 레이어가 스태킹 모델의 조합으로 구성되며 계층의 수를 증가시켜 각 계층의 오분류율을 최소화하여 성능을 개선한다. 4가지 유형의 데이터셋을 이용한 실험을 통해 제안 기법이 기존 기법에 비해 분류 성능이 우수함을 보인다.

• Journal of Information Processing Systems
• /
• 제19권5호
• /
• pp.652-662
• /
• 2023

Identifying highly discriminating genes is a critical step in tumor recognition tasks based on microarray gene expression profile data and machine learning. Gene selection based on tree models has been the subject of several studies. However, these methods are based on a single-tree model, often not robust to ultra-highdimensional microarray datasets, resulting in the loss of useful information and unsatisfactory classification accuracy. Motivated by the limitations of single-tree-based gene selection, in this study, ensemble gene selection methods based on multiple-tree models were studied to improve the classification performance of tumor identification. Specifically, we selected the three most representative tree models: ID3, random forest, and gradient boosting decision tree. Each tree model selects top-n genes from the microarray dataset based on its intrinsic mechanism. Subsequently, three ensemble gene selection methods were investigated, namely multipletree model intersection, multiple-tree module union, and multiple-tree module cross-union, were investigated. Experimental results on five benchmark public microarray gene expression datasets proved that the multiple tree module union is significantly superior to gene selection based on a single tree model and other competitive gene selection methods in classification accuracy.

• 인터넷정보학회논문지
• /
• 제17권2호
• /
• pp.49-57
• /
• 2016

앙상블 학습 기법은 개별 모형보다 더 좋은 예측 성과를 얻기 위해 다수의 분류기를 결합하는 것으로 예측 성과를 향상시키는데에 매우 유용한 것으로 알려져 있다. 배깅은 단일 분류기의 예측 성과를 향상시키는 대표적인 앙상블 기법중의 하나이다. 배깅은 원 학습 데이터로부터 부트스트랩 샘플링 방법을 통해 서로 다른 학습 데이터를 추출하고, 각각의 부트스트랩 샘플에 대해 학습 알고리즘을 적용하여 서로 다른 다수의 기저 분류기들을 생성시키게 되며, 최종적으로 서로 다른 분류기로부터 나온 결과를 결합하게 된다. 배깅에서 부트스트랩 샘플은 원 학습 데이터로부터 램덤하게 추출한 샘플로 각각의 부트스트랩 샘플이 동일한 정보를 가지고 있지는 않으며 이로 인해 배깅 모형의 성과는 편차가 발생하게 된다. 본 논문에서는 이와 같은 부트스트랩 샘플을 최적화함으로써 표준 배깅 앙상블의 성과를 개선시키는 새로운 방법을 제안하였다. 제안한 모형에서는 앙상블 모형의 성과를 개선시키기 위해 부트스트랩 샘플링을 최적화하였으며 이를 위해 유전자 알고리즘이 활용되었다. 본 논문에서는 제안한 모형을 국내 부도 예측 문제에 적용해 보았으며, 실험 결과 제안한 모형이 우수한 성과를 보였다.