• Journal of IKEEE
• /
• v.24 no.2
• /
• pp.475-485
• /
• 2020

Recently, a crowdsensing system that provides a new sensing service with real-time sensing data provided from a user's device including a sensor without installing a separate sensor has attracted attention. In the crowdsensing system, meaningless data may be provided due to a user's operation error or communication problem, or false data may be provided to obtain compensation. Therefore, the detection and removal of the abnormal data determines the quality of the crowdsensing service. The proposed methods in the past to detect these anomalies are not efficient for the fast-changing environment of crowdsensing. This paper proposes an anomaly data detection method by extracting the characteristics of continuously and rapidly changing sensing data environment by using machine learning technology and modeling it with an appropriate algorithm. We show the performance and feasibility of the proposed system using deep learning binary classification model of supervised learning and autoencoder model of unsupervised learning.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.4
• /
• pp.91-98
• /
• 2022

Lifelog data continuously collected through a wearable device may contain many outliers, so in order to improve data quality, it is necessary to find and remove outliers. In general, since the number of outliers is less than the number of normal data, a class imbalance problem occurs. To solve this imbalance problem, we propose a method that applies Variational AutoEncoder to outliers. After preprocessing the outlier data with proposed method, it is verified through a number of machine learning models(classification). As a result of verification using body weight data, it was confirmed that the performance was improved in all classification models. Based on the experimental results, when analyzing lifelog body weight data, we propose to apply the LightGBM model with the best performance after preprocessing the data using the outlier processing method proposed in this study.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2015.04a
• /
• pp.702-705
• /
• 2015

본 논문에서는 하듐 환경에서 시스템 정보의 이상탐지를 위한 시각화 기능을 설계 및 구현한다. 제안한 이상탐지 시각화 기능은 크게 세 단계로 구분된다. 먼저, 각 노드로부터 시스템 로그 데이터(캐시 및 메인 메모리)를 수집하여 하이브(Hive) 저장한다. 그리고 저장한 데이터에 3-시그마 규칙을 적용하여 이상탐지를 수행한 후 관계형 데이터베이스에 적합하도록 재가공한다. 마지막으로, 스쿱(Sqoop)을 통해 RDBMS(MariaDB)에 이상탕지 결과를 저장하고, DHTMLX 차트 라이브러리를 사용하여 이를 시각화한다. 시각화 결과, 로그 데이터의 이상탐지와 데이터간의 상관관계를 직관적으로 이해할 수 있게 되었다.

• Journal of Internet Computing and Services
• /
• v.18 no.1
• /
• pp.27-35
• /
• 2017

In near future, IoT data is expected to be a major portion of Big Data. Moreover, sensor data is expected to be major portion of IoT data, and its' research is actively carried out currently. However, processed results may not be trusted and used if outlier data is included in the processing of sensor data. Therefore, method for detection and deletion of those outlier data before processing is studied in this paper. Moreover, we used Spark which is memory based distributed processing environment for fast processing of big sensor data. The detection and deletion of outlier data consist of four stages, and each stage is implemented with Mapper and Reducer operation. The proposed method is compared in three different processing environments, and it is expected that the outlier detection and deletion performance is best in the distributed Spark environment as data volume is increasing.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.2
• /
• pp.199-206
• /
• 2022

A malfunction or breakdown of a manufacturing facility leads to product defects and the suspension of production lines, resulting in huge financial losses for manufacturers. Due to the spread of smart factory services, a large amount of data is being collected in factories, and AI-based research is being conducted to predict and diagnose manufacturing facility breakdowns or manufacturing site efficiency. However, because of the characteristics of manufacturing data, such as a severe class imbalance about abnormalities and ambiguous label information that distinguishes abnormalities, developing classification or anomaly detection models is highly difficult. In this paper, we present an deep learning algorithm for anomaly detection of a manufacturing facility using reconstruction loss of CNN-based model and ananlyze its performance. The algorithm detects anomalies by relying solely on normal data from the facility's manufacturing data in the exclusion of abnormal data.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2021.07a
• /
• pp.5-6
• /
• 2021

본 논문에서는 Semi-supervised Learning 방식의 이상감지 방법을 제안한다. 취득한 소음 데이터를 이미지화 시킨 후 Convolution AutoEncoder 학습 방법을 이용하여 모델을 학습한다. 고장 데이터와 정상 데이터 간의 데이터 불균형 문제가 대두되기 때문에 정상 데이터만을 활용한 이상감지는 실제 산업현장의 상황에 알맞게 사용할 수 있을 것이라 기대한다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.4
• /
• pp.647-660
• /
• 2022

Recent works demonstrate that the semi-supervised anomaly detection method functions quite well in the environment with normal data and some anomalous data. However, abnormal data shortages can occur in an environment where it is difficult to reserve anomalous data, such as an unknown attack in the cyber security fields. In this paper, we propose ADA-PH(Abnormal Data Augmentation Method using Perturbation based on Hypersphere), a novel anomalous data augmentation method that is applicable in an environment where abnormal data is insufficient to secure the performance of the semi-supervised anomaly detection method. ADA-PH generates abnormal data by perturbing samples located relatively far from the center of the hypersphere. With the network intrusion detection datasets where abnormal data is rare, ADA-PH shows 23.63% higher AUC performance than anomaly detection without data augmentation and even performs better than the other augmentation methods. Also, we further conduct quantitative and qualitative analysis on whether generated abnormal data is anomalous.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2017.05a
• /
• pp.172-172
• /
• 2017

본 연구는 차량의 AW(AutoWiping) 기능을 위해 장착된 강우센서를 이용하여 강우정보를 생산하는 기술을 개발하고자 하였다. AW(AutoWiping) 기능이란 차량 앞창(Windshield)에 빗방울이 맺히게 되면 광신호의 산란으로 인해 수광부에 들어오는 감소되는 광신호의 정도에 따라 차량 와이퍼의 속도를 결정해 주는 기능이다. 빗방울이 많이 맺힐수록 광신호는 감소되며 와이퍼는 더 빠른 속도로 작동을 하게 된다. 여기서 강우센서가 강우량이 많으면 감소된 광신호 데이터를 표출하는 현상을 이용하여 강우정보를 생산한다. 강우센서는 총 8개의 채널로 이루어져있고, 초당 250개의 광신호 데이터를 수집하며, 10분이면 약 120만 개의 데이터가 생산되게 된다. 이 대량의 데이터에서 정확한 강우량을 산출하기 위해 강우센서의 초기값과 와이퍼 이동시 발생하는 순간 이상치를 제거해야 한다. 하지만 일일이 수백만 개 이상의 데이터에서 모든 이상치를 제거하는 작업은 불가능하다. 따라서 이상치를 포함한 회귀 분석 방법을 연구하였고, 인공강우 발생기를 이용하여 광신호를 강우량으로 환산하는 2가지 회귀식이 유도되었다. 이들은 각각 이상치를 모두 포함시켜 독립변수(광신호)에 따라 종속변수(강우량)의 값이 변화하는 관계를 나타내는 선형회귀분석(model 1), 임계치를 정하여 일정 이상치가 제거된 신호만 통과시키는 대역통과 필터링 분석(model 2)으로 유도된 회귀식을 실강우에 회귀식을 적용하여 정확도를 분석하였다.

• 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.

• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.229-247
• /
• 2023

With the increasing emphasis on anomaly detection across various fields, diverse anomaly detection algorithms have been developed for various data types and anomaly patterns. However, the performance of anomaly detection algorithms is generally evaluated on publicly available datasets, and the specific performance of each algorithm on anomalies of particular types remains unexplored. Consequently, selecting an appropriate anomaly detection algorithm for specific analytical contexts poses challenges. Therefore, in this paper, we aim to investigate the types of anomalies and various attributes of data. Subsequently, we intend to propose approaches that can assist in the selection of appropriate anomaly detection algorithms based on this understanding. Specifically, this study compares the performance of anomaly detection algorithms for four types of anomalies: local, global, contextual, and clustered anomalies. Through further analysis, the impact of label availability, data quantity, and dimensionality on algorithm performance is examined. Experimental results demonstrate that the most effective algorithm varies depending on the type of anomaly, and certain algorithms exhibit stable performance even in the absence of anomaly-specific information. Furthermore, in some types of anomalies, the performance of unsupervised anomaly detection algorithms was observed to be lower than that of supervised and semi-supervised learning algorithms. Lastly, we found that the performance of most algorithms is more strongly influenced by the type of anomalies when the data quantity is relatively scarce or abundant. Additionally, in cases of higher dimensionality, it was noted that excellent performance was exhibited in detecting local and global anomalies, while lower performance was observed for clustered anomaly types.