• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.11
• /
• pp.435-442
• /
• 2018

Outlier detection means to detect data samples that deviate significantly from the distribution of normal data. Most outlier detection methods calculate an outlier score that indicates the extent to which a data sample is out of normal state and determine it to be an outlier when its outlier score is above a given threshold. However, since the range of an outlier score is different for each data and the outliers exist at a smaller ratio than the normal data, it is very difficult to determine the threshold value for an outlier score. Further, in an actual situation, it is not easy to acquire data including a sufficient amount of outliers available for learning. In this paper, we propose a clustering-based outlier detection method by constructing a model representing a normal data region using only normal data and performing binary classification of outliers and normal data for new data samples. Then, by dividing the given normal data into chunks, and constructing a clustering model for each chunk, we expand it to the ensemble method combining the decision by the models and apply it to the streaming data with dynamic changes. Experimental results using real data and artificial data show high performance of the proposed method.

• Communications for Statistical Applications and Methods
• /
• v.18 no.4
• /
• pp.517-525
• /
• 2011

This paper shows the performance evaluation of a robust estimator based on the GARCH model. We first introduce the method of a robust estimate in the GARCH model and the method of an outlier detection in the GARCH model. The results of the real internet traffic data show the out-performance of the robust estimator over the outlier detection method in the GARCH model. In addition, the method of the robust estimate is less complex than the method of the outlier detection method in the GARCH model.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2016.05a
• /
• pp.289-289
• /
• 2016

신뢰성 높은 홍수관리는 경향성 분석, 이상치 판정 등의 전처리를 수행한 입력 자료를 구축하는 것을 필요로 한다. 경향성 분석은 방법에 따라 경향성의 유무가 다르게 나타나기 때문에 하나의 방법으로만 판단하기 어려우며, 이상치 분석은 지역 특성에 따라 기준이 변동하므로 일정한 기준을 적용하기가 어려워 주로 수동으로 이루어지며 이 작업을 완료하는 데에는 많은 시간이 소요된다. 입력 자료 전처리에 수반되는 비용과 시간을 절감하기 위해 이러한 문제점의 개선이 필요한 실정이다. 따라서 본 연구의 목적은 농촌유역 홍수관리를 위한 자료처리 요소 모듈을 개발하는 데 있다. 홍수관리를 위한 자료처리 요소 모듈은 크게 기상자료의 경향성을 분석하는 모듈과 수위자료의 이상치를 탐지하고 판정하는 모듈로 구성하였다. 경향성 분석 모듈은 모수적 방법인 t-test와 비모수적 방법인 Hotelling-Pabst test 및 Mann-Kendall test를 분석 방법으로 제공하여 하나의 입력 자료로 세 가지 방법으로 분석한 결과를 비교할 수 있도록 개발하였다. 이상치 탐지 모듈은 IQR (interquartile range) 규칙과 규칙기반의 방법을 이용한 이상치 탐지를 제공할 수 있도록 개발하였다. 개발된 모듈은 한강 유역의 용당저수지에 적용하여 검정을 실시하였다. 본 연구에서 개발된 농촌유역 홍수관리를 위한 자료처리 요소 모듈은 추후 홍수관리 및 그에 관한 연구를 하는데 있어 활용될 수 있을 것으로 기대된다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.1208-1210
• /
• 2023

이상치는 주로 저빈도로 발생하기 때문에, 이상치 탐지 분야에서는 정상 데이터만을 이용한 비지도 기반 학습 모델을 사용하는 방법들이 제안되었다. 따라서, 본 논문에서는 반려동물 센서 데이터를 이용해 비지도 기반 모델인 DASVDD을 활용하여 이상치를 탐지한다. 하지만 데이터셋에 이상치가 존재하지 않아 반려동물이 고빈도로 보여주는 A행동군(서다, 앉다, 엎드리다, 눕다, 걷다), 저빈도로 보여주는 B행동군(킁킁대다, 먹다)으로 분리하여 학습을 진행한다. 모델의 성능은 ROC-AUC을 기준으로 79.05%의 성능을 보여주는 것을 확인하였다.

• The Korean Journal of Applied Statistics
• /
• v.29 no.3
• /
• pp.399-408
• /
• 2016

There is a need to detect heteroscedasticity in a regression analysis; however, it invalidates the standard inference procedure. The diagnostics on heteroscedasticity may be distorted when both outliers and heteroscedasticity exist. Available heteroscedasticity detection methods in the presence of outliers usually use robust estimators or separating outliers from the data. Several approaches have been suggested to identify outliers in the heteroscedasticity problem. In this article conventional tests on heteroscedasticity are modified by using a sequential outlier detection methods to separate outliers from contaminated data. The performance of the proposed method is compared with original tests by a Monte Carlo study and examples.

• The Korean Journal of Applied Statistics
• /
• v.24 no.5
• /
• pp.979-986
• /
• 2011

A linear regression method is commonly used to analyze data because of its simplicity and applicability; however, it is well known that data may contain some outliers and influential cases that may have a harmful effect on a statistical analysis. Thus detection and examination of outliers or influential cases are important parts of data analysis. In detecting multiple outliers, masking effects usually occur and make it difficult to identify the true outliers. We propose to use dynamic plots as a method resistant to masking effect. The procedure using dynamic plots is useful to find appropriate basic sets with which a dependent outliers detection method start and detect a true outliers set. Examples are given to demonstrate the effectiveness of the suggested idea.

• Journal of Internet Computing and Services
• /
• v.22 no.3
• /
• pp.45-52
• /
• 2021

Although many studies have been conducted to identify unknown attacks in cyber security intrusion detection systems, studies based on outliers are attracting attention. Accordingly, we identify outliers by defining categories for unknown attacks. The unknown attacks were investigated in two categories: first, there are factors that generate variant attacks, and second, studies that classify them into new types. We have conducted outlier studies that can identify similar data, such as variants, in the category of studies that generate variant attacks. The big problem of identifying anomalies in the intrusion detection system is that normal and aggressive behavior share the same space. For this, we applied a technique that can be divided into clear types for normal and attack by discrete wavelet transformation and detected anomalies. As a result, we confirmed that the outliers can be identified through One-Class SVM in the data reconstructed by discrete wavelet transform.

• Journal of Korean Society of Transportation
• /
• v.35 no.6
• /
• pp.469-480
• /
• 2017

On April 1, 2004, KTX (Korea Train eXpress), the first HSR (High-Speed Rail) in Korea, was introduced to Gyeongbu Line. The introduction of the KTX service led to a change in the number of passengers for Gyeongbu Line. Previous studies have analyzed the pre and post-event changes of the intervening events by either simple statistics or intervention ARIMA analysis. However, the intervention ARIMA model has a limitation that several assumptions such as the occurrence time and the type of intervention events are necessary. To this end, this study analyzed the effects of intervention event on the number of passengers using the Gyeongbu line based on a time series outlier detection technique which can overcome limitations in the previous studies. The time series outlier detection technique can analyze the time, effect type and size of an intervention event without the assumption of the time and effect type of the intervention event. The data were collected from the Korea Transport Database (KTDB) for twelve years from 2003 to 2014 (144 months). The analysis results showed that the size of the influence type in the same intervention events was different across the major city routes, and the intervention event which could not be found by previous study methods was also found.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2016.07a
• /
• pp.73-74
• /
• 2016

IoT 데이터는 비정형 데이터로 가공되고 분석하였을 때 비로소 가치를 갖기에 전 세계적으로 빅데이터 기술에 관심이 집중되고 있다. IoT 데이터 중 많은 부분을 차치하는 센서 데이터는 수집이 용이하고 활용범위가 넓기 때문에 여러 분야에서 사용되고 있다. 하지만 센서가 정상적으로 작동하지 못한 경우에는 실제와는 다른 값인 이상치를 포함하여 왜곡된 결과가 도출되어 활용할 수 없는 경우가 생긴다. 따라서 본 논문에서는 정확한 결과를 도출하기 위하여 수집된 원자료의 데이터를 분석하기 전에 이상치 탐지 및 제거를 하고자 한다. 또한 점점 늘어나고 있는 대용량 데이터를 신속하게 처리하기 위하여 메모리 접근방식인 스파크를 사용한 분산처리환경에서 이상치 탐지 및 제거하는 것을 제안한다. 맵리듀스 기반의 이상치 탐지 및 제거는 총 4단계로 나누어 구현하였으며 제안한 기법의 성능 평가를 위해 총 3가지 환경에서 비교하여 실험하였다. 실험을 통해 데이터의 용량이 커질수록 분산처리환경에서 스파크를 사용하여 처리하는 방식이 가장 빠를 것 이라는 결과를 얻었다.

• KIISE Transactions on Computing Practices
• /
• v.23 no.2
• /
• pp.128-133
• /
• 2017

In recent years, the number of systems for the analysis of large volumes of data is increasing. Hadoop, a representative big data system, stores and processes the large data in the distributed environment of multiple servers, where system-resource management is very important. The authors attempted to detect anomalies from the rapid changing of the log data that are collected from the multiple servers using simple but efficient anomaly-detection techniques. Accordingly, an Apache Hive storage architecture was designed to store the log data that were collected from the multiple servers in the Hadoop ecosystem. Also, three anomaly-detection techniques were designed based on the moving-average and 3-sigma concepts. It was finally confirmed that all three of the techniques detected the abnormal intervals correctly, while the weighted anomaly-detection technique is more precise than the basic techniques. These results show an excellent approach for the detection of log-data anomalies with the use of simple techniques in the Hadoop ecosystem.