• 한국멀티미디어학회논문지
• /
• 제23권6호
• /
• pp.747-755
• /
• 2020

In this paper, we deal with outlier data problems that occur when constructing a PM_2.5 fine dust forecasting system using a neural network. In general, when learning a neural network, some of the data are not helpful for learning, but rather disturbing. Those are called outlier data. When they are included in the training data, various problems such as overfitting occur. In building a PM_2.5 fine dust concentration forecasting system using neural network, we have found several outlier data in the training data. We, therefore, remove them, and then make learning 3 ways. Over_outlier model removes outlier data that target concentration is low, but the model forecast is high. Under_outlier model removes outliers data that target concentration is high, but the model forecast is low. All_outlier model removes both Over_outlier and Under_outlier data. We compare 3 models with a conventional outlier removal model and non-removal model. Our outlier removal model shows better performance than the others.

• 한국멀티미디어학회논문지
• /
• 제23권8호
• /
• pp.915-926
• /
• 2020

Outlier detection methods help one to identify unusual instances in data that may correspond to erroneous, exceptional, or surprising events or behaviors. This work studies conditional outlier detection, a special instance of the outlier detection problem, in the context of incorrect data label identification. Unlike conventional (unconditional) outlier detection methods that seek abnormalities across all data attributes, conditional outlier detection assumes data are given in pairs of input (condition) and output (response or label). Accordingly, the goal of conditional outlier detection is to identify incorrect or unusual output assignments considering their input as condition. As a solution to conditional outlier detection, this paper proposes the ratio-based outlier scoring (ROS) approach and its variant. The propose solutions work by adopting conventional outlier scores and are able to apply them to identify conditional outliers in data. Experiments on synthetic and real-world image datasets are conducted to demonstrate the benefits and advantages of the proposed approaches.

• 한국멀티미디어학회논문지
• /
• 제25권6호
• /
• pp.886-893
• /
• 2022

Outlier detection refers to the task of detecting data that deviate significantly from the normal data distribution. Most outlier detection methods compute an outlier score which indicates the degree to which a data sample deviates from normal. However, setting a threshold for an outlier score to determine if a data sample is outlier or normal is not trivial. In this paper, we propose a binary prediction method for outlier detection based on spectral clustering and one-class SVM ensemble. Given training data consisting of normal data samples, a clustering method is performed to find clusters in the training data, and the ensemble of one-class SVM models trained on each cluster finds the boundaries of the normal data. We show how to obtain a threshold for transforming outlier scores computed from the ensemble of one-class SVM models into binary predictive values. Experimental results with high dimensional text data show that the proposed method can be effectively applied to high dimensional data, especially when the normal training data consists of different shapes and densities of clusters.

• 한국멀티미디어학회논문지
• /
• 제24권9호
• /
• pp.1242-1250
• /
• 2021

Detection of outliers deviating normal data distribution in high dimensional data is an important technique in many application areas. In this paper, a distance-based outlier detection method using landmarks in high dimensional data is proposed. Given normal training data, the k-means clustering method is applied for the training data in order to extract the centers of the clusters as landmarks which represent normal data distribution. For a test data sample, the distance to the nearest landmark gives the outlier score. In the experiments using high dimensional data such as images and documents, it was shown that the proposed method based on the landmarks of one-tenth of training data can give the comparable outlier detection performance while reducing the time complexity greatly in the testing stage.

• 한국컴퓨터정보학회논문지
• /
• 제21권8호
• /
• pp.77-84
• /
• 2016

In general statistical analysis, we need to make a normal assumption. If this assumption is not satisfied, we cannot expect a good result of statistical data analysis. Most of statistical methods processing the outlier and noise also need to the assumption. But the assumption is not satisfied in big data because of its large volume and heterogeneity. So we propose a methodology based on box-plot and data smoothing for controling outlier and noise in big data analysis. The proposed methodology is not dependent upon the normal assumption. In addition, we select patent documents as target domain of big data because patent big data analysis is a important issue in management of technology. We analyze patent documents using big data learning methods for technology analysis. The collected patent data from patent databases on the world are preprocessed and analyzed by text mining and statistics. But the most researches about patent big data analysis did not consider the outlier and noise problem. This problem decreases the accuracy of prediction and increases the variance of parameter estimation. In this paper, we check the existence of the outlier and noise in patent big data. To know whether the outlier is or not in the patent big data, we use box-plot and smoothing visualization. We use the patent documents related to three dimensional printing technology to illustrate how the proposed methodology can be used for finding the existence of noise in the searched patent big data.

• 한국농공학회논문집
• /
• 제61권1호
• /
• pp.107-120
• /
• 2019

Reservoir water level data identify the current water storage of the reservoir, and they are utilized as primary data for management and research of agricultural water. For the reservoir storage management, Korea Rural Community Corporation (KRC) installed water level stations at around 1,600 agricultural reservoirs and has been collecting the water level data every 10 minutes. However, various kinds of outliers due to noise and erroneous problems are frequently appearing because of environmental and physical causes. Therefore, it is necessary to detect outlier and improve the quality of reservoir water level data to utilize the water level data in purpose. This study was conducted to detect and classify outlier and normal data using two different models including the threshold model and the artificial neural network (ANN) model. The results were compared to evaluate the performance of the models. The threshold model identifies the outlier by setting the upper/lower bound of water level data and variation data and by setting bandwidth of water level data as a threshold of regarding erroneous water level. The ANN model was trained with prepared training dataset as normal data (T) and outlier (F), and the ANN model operated for identifying the outlier. The models are evaluated with reference data which were collected reservoir water level data in daily by KRC. The outlier detection performance of the threshold model was better than the ANN model, but ANN model showed better detection performance for not classifying normal data as outlier.

• ETRI Journal
• /
• 제41권4호
• /
• pp.437-451
• /
• 2019

Outlier detection techniques play an important role in enhancing the reliability of data communication in wireless sensor networks (WSNs). Considering the importance of outlier detection in WSNs, many outlier detection techniques have been proposed. Unfortunately, most of these techniques still have some potential limitations, that is, (a) high rate of false positives, (b) high time complexity, and (c) failure to detect outliers online. Moreover, these approaches mainly focus on either temporal outliers or spatial outliers. Therefore, this paper aims to introduce novel algorithms that successfully detect both temporal outliers and spatial outliers. Our contributions are twofold: (i) modifying the Hampel Identifier (HI) algorithm to achieve high accuracy identification rate in temporal outlier detection, (ii) combining the Gaussian process (GP) model and graph-based outlier detection technique to improve the performance of the algorithm in spatial outlier detection. The results demonstrate that our techniques outperform the state-of-the-art methods in terms of accuracy and work well with various data types.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2008년도 International Symposium on Remote Sensing
• /
• pp.410-413
• /
• 2008

Existing Outlier detections have been widely studied in geosensor networks. Recently, machine learning and data mining have been applied the outlier detection method to build a model that distinguishes outliers based on anchored criterion. However, it is difficult for the existing methods to detect outliers against incoming time-variant data, because outlier detection needs to monitor incoming data and classify irregular attacks. Therefore, in order to solve the problem, we propose a time-variant outlier detection using 2-dimensional grid method based on unanchored criterion. In the paper, outliers using geosensor data was performed to classify efficiently. The proposed method can be utilized applications such as network intrusion detection, stock market analysis, and error data detection in bank account.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제7권11호
• /
• pp.435-442
• /
• 2018

이상치 탐지는 정상 데이터 분포를 크게 벗어나는 데이터 샘플을 탐지하는 것을 의미한다. 대부분의 이상치 탐지 방법은 데이터 샘플이 정상 상태를 벗어나는 정도를 나타내는 이상치 지수(outlier score)를 계산하여 주어진 임계값 이상일 때 이상치로 판정한다. 그러나, 데이터마다 이상치 지수의 범위가 다양하고 정상 데이터에 비해 이상치 데이터는 적은 비율로 존재하기 때문에 이상치 지수에 대한 임계값을 결정하기는 매우 어렵다. 또한, 실제 상황에서는 학습에 이용할 수 있는 충분한 양의 이상치를 포함하는 데이터의 획득이 용이하지 않다. 본 논문에서는 정상 데이터가 주어졌을 때 이를 이용하여 정상 데이터 영역을 나타내는 모델을 구성하고 새로운 데이터 샘플에 대해 이상치와 정상치의 이진 분류를 수행하는 방법으로 군집화 기반 이상치 탐지 방법을 제안한다. 그리고, 주어진 정상 데이터를 청크로 나누고 각 청크에 대해 클러스터링 모델을 구성한 후 모델들에 의한 이상치 판정 결과를 결합하는 앙상블 방법과 동적 변화가 있는 스트리밍 데이터에서의 적용 방법으로 확장한다. 실제 데이터와 인공 데이터를 이용한 실험결과는 제안 방법의 높은 성능을 보여준다.

• 한국지반환경공학회 논문집
• /
• 제16권5호
• /
• pp.43-53
• /
• 2015

본 연구에서는 빅데이터 범주에 포함되는 각종 계측 데이터를 대상으로 각종 이상치를 판단하기 위한 기법을 고안하고, 인공 데이터 및 실 계측 데이터를 이용한 이상치 분석을 수행하였다. 계측결과에 대한 1차 차분 값 및 오차율을 적용한 규칙기반 방법은 큰 규모의 Short fault 분석 및 일정 기간 계측값에 변화가 발생하지 않는 경우의 Constant fault 분석에 효과적으로 적용될 수 있었으나, 독립적인 단일 데이터셋만을 이용하는 관계로 큰 변화폭을 보이는 실 계측 데이터의 정상 데이터를 이상치로 오판하는 문제점이 있었다. 규칙기반 방법을 이용한 Noise fault 분석은 적정 데이터 윈도우 사이즈의 선택 및 이상치 판정용 한계값 선정상의 문제로 인해 실 계측 데이터 적용에 한계가 있었다. 이종 데이터 간 상관분석 방법은 학습 데이터의 적정범위 선정이 선행된다면 장단기 계측 데이터의 이상 거동 및 국부적 이상치 판정에 매우 효과적으로 이용될 수 있음을 알 수 있었다.