• The Korean Journal of Applied Statistics
• /
• v.33 no.3
• /
• pp.257-267
• /
• 2020

We introduce multivariate exponential smoothing models based on a vector innovations structural time series framework. The models enable us to exploit potential inter-series dependencies to improve the fit and forecasts of multivariate (vector) time series. Models are applied to forecast the exchange rates of the UK pound (UKP) and US dollar (USD) against the Korean won (KRW) observed on monthly basis; subseqently, we compare their performance with alternative models. We observe that the multivariate exponential smoothing models are superior to alternatives.

• The Korean Journal of Applied Statistics
• /
• v.33 no.6
• /
• pp.723-738
• /
• 2020

In this paper, we study a method to discriminate between trend stationary and difference stationary processes. Since a crucial ingredient of this discrimination is to determine the existence of unit root, we can use a unit root testing strategy. So, we introduce a discrimination based on unit root testing and propose the method using the adaptive lasso. Our Monte Carlo simulation experiments show that the adaptive lasso improves the discrimination accuracy when the process is trend stationary, but has lower accuracy than unit root strategy where the process is difference stationary.

• The Korean Journal of Applied Statistics
• /
• v.29 no.5
• /
• pp.907-920
• /
• 2016

This study suggests an outlier detection algorithm that uses quantile autoregressive model in time series data, eventually applying it to actual stock manipulation cases by comparing its performance to existing methods. Studies on outlier detection have traditionally been conducted mostly in general data and those in time series data are insufficient. They have also been limited to a parametric model, which is not convenient as it is complicated with an analysis that takes a long time. Thus, we suggest a new algorithm of outlier detection in time series data and through various simulations, compare it to existing algorithms. Especially, the outlier detection algorithm in time series data can be useful in finding stock manipulation. If stock price which had a certain pattern goes out of flow and generates an outlier, it can be due to intentional intervention and manipulation. We examined how fast the model can detect stock manipulations by applying it to actual stock manipulation cases.

• Journal of Astronomy and Space Sciences
• /
• v.25 no.4
• /
• pp.395-404
• /
• 2008

We explore the linkage between the long term variability of the Sun and earth's climate change by analysing periodicities of time series of solar proxies and global temperature anomalies. We apply the power spectral estimation method named as the periodgram to solar proxies and global temperature anomalies. We also decompose global temperature anomalies and reconstructed total solar irradiance into each local variability components by applying the EMD (Empirical Mode Decomposition) and MODWT MRA (Maximal Overlap Discrete Wavelet Multi Resolution Analysis). Powers for solar proxies at low frequencies are lower than those of high frequencies. On the other hand, powers for temperature anomalies show the other way. We fail to decompose components which having lager than 40 year variabilities from EMD, but both residuals are well decomposed respectively. We determine solar induced components from the time series of temperature anomalies and obtain 39% solar contribution on the recent global warming. We discuss the climate system can be approximated with the second order differential equation since the climate sensitivity can only determine the output amplitude of the signal.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.1
• /
• pp.1-12
• /
• 2023

It is very time-intensive to obtain data with labels on anomaly detection tasks for multivariate time series. Therefore, several studies have been conducted on unsupervised learning that does not require any labels. However, a well-done integrative survey has not been conducted on in-depth discussion of learning architecture and property for multivariate time series anomaly detection. This study aims to explore the characteristic of well-known architectures in anomaly detection of multivariate time series. Additionally, architecture was categorized by using top-down and bottom-up approaches. In order toconsider real-world anomaly detection situation, we trained models with dataset such as power grids or Cyber Physical Systems that contains realistic anomalies. From experimental results, we compared and analyzed the comprehensive performance of each architecture. Quantitative performance were measured using precision, recall, and F1 scores.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10a
• /
• pp.334-336
• /
• 1999

본 논문에서는 시계열 데이터베이스에서 임의 계수의 이동평균 변환을 지원하는 서브시퀀스 매칭 알고리즘을 제안한다. 응용분야와 분석하려고 하는 시계열 데이터의 특성에 따라 잡음의 영향을 줄이는 정도와 경향을 파악하는 주기가 달라지므로 이동평균 계수의 선택도 달라진다. 본 논문에서는 하나의 이동평균 계수에 대해서 생성한 인덱스만을 이용하여 인덱스가 생성되어 있지 않은 계수에 대해서도 탐색을 수행하는 방법을 제안한다. 이때, 제안된 탐색 기법이 질의 결과로 반환되어야 할 서브시퀀스를 모두 찾아내지 못하는 착오 기각이 발생하지 않음을 증명한다. 실험 결과, 모든 이동평균 계수에 대해 인덱스가 생성되어 있는 경우와 비교하여 탐색 성능의 저하는 42%이내였으며, 제안된 알고리즘의 탐색 성능이 순차 검색에 비하여 초대 2.7배 우수하였다.

• Annual Conference of KIPS
• /
• 2013.05a
• /
• pp.250-253
• /
• 2013

최근 카메라 영상을 이용한 제스처 인식 관련 연구가 활발히 진행되고 있다. 카메라 영상을 이용한 제스처 인식에서 많이 사용되는 학습 알고리즘에는 확률 그래프 모델인 HMM과 CRF 등이 있다. 이 학습 알고리즘들은 다차원의 연속된 실수 데이터를 가지고 모델을 학습하면 계산량이 많아진다. 본 논문에서는 팔 관절 위치 데이터를 k-평균 군집화 과정을 거쳐 1차원의 시계열 데이터로 변환 후, 제스처별로 HMM 모델을 학습하는 방법을 제안한다. 키넥트 센서를 통해 얻은 팔 관절 위치 데이터에 k-평균 군집화를 적용하여 1차원 시계열 데이터를 생성하고, 이를 HMM의 학습 및 인식에 사용한다. 본 논문에서 제안하는 방법의 성능을 분석하기 위하여, 다른 시계열 학습 알고리즘인 AP+DTW를 이용한 방법과의 비교 실험을 포함해 다양한 실험들을 수행하였다.

• Journal of Korea Water Resources Association
• /
• v.49 no.3
• /
• pp.177-186
• /
• 2016

Drought is a water-related natural disaster which can be simply described as spatially and temporally sequential absence of water. However, its characteristics are very difficult to define. For this reason, the preparation and mitigation from drought events have not been successful. In the current study, we illustrated a design drought estimation approach of water resources infrastructures as well as the existing theoretical one to prepare and mitigate drought disasters. Theoretical and simulation methods were tested including three time series models such as autoregressive (AR), Gamma AR, Copula AR models. The results indicated that for South Korea region, the simulation-based method to estimate drought frequency presented better performance and all the three time series models show similar performance to each other. The current drought event occurring in South Korea was investigated with dividing South Korea into four basins as Han River, Nakdong River, Geum River, and Nakdong River basins. The results showed that two middle and north basins presented significant drought events with 3 year drought duration and around 40 year return period while the other two southern regions illustrated relatively weaker drought events.

• Journal of the Korean Data and Information Science Society
• /
• v.25 no.4
• /
• pp.807-817
• /
• 2014

This paper considers multivariate time series modelling of PM10 data in Korea collected from 2008 to 2011. We consider both temporal and spatial dependencies of PM10 by applying the sparse vector autoregressive (sVAR) modelling proposed by Davis et al. (2013). It utilizes the partial spectral coherence to measure cross correlation between different regions, in turn provides the sparsity in the model while balancing the parsimony of model and the goodness of fit. It is also shown that sVAR performs better than usual vector autoregressive model (VAR) in forecasting.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2015.05a
• /
• pp.462-462
• /
• 2015

이 논문에서는 시 공간적 토양수분 변화를 파악하기 위해 다년간 축적된 실측 토양수분 데이터를 이용하여 단변량 시계열 분석을 하였다. 지형에 따른 토양수분 변화를 알아보기 위해 경기도 파주에 위치한 설마천 유역의 산지사면 중 한 단면을 선정하였으며, 깊이에 따른 변동성은 깊이 10cm와 30cm에서 측정한 토양수분 데이터를 이용하여 분석하였다. 또한, 연도별 토양수분의 변화를 파악하고 토양수분을 예측하기 위해 2010-2013년의 토양수분 데이터를 일단위로 단변량 모델링을 시도하였다. 그 결과, 연도별 변화에 따른 경향성은 보이지 않았으며 대부분의 지점에서 ARMA(1, 1) 또는 ARMA(1, 0) 모형으로 모의되었다. 2시간 간격의 1-2개월 단기간 토양수분 데이터를 모의한 선행연구와 달리 본 연구에서는 낮은 차수의 모형을 보였다. 지형적 토양수분 거동을 살펴보면 상부사면에 위치하고 있는 지점에서는 모두 ARMA(1, 1)로 표현되지만 하부사면에 위치한 지점들은 연도나 심도에 따라 ARMA(1, 0)으로 모의된다. 단변량 모형의 정확도를 알아보기 위해 R2와 RMSE를 비교하였다. 10cm 깊이에서는 경향성을 보이지 않으나, 30cm 깊이에서는 사면하부로 갈수록 R2는 작아지고 RMSE는 커져, 하부사면에서의 모델링이 상부사면에 비해 정확도가 낮음을 보였다. 또한 2012년 토양수분 자료를 이용하여 2013년 토양수분을 예측하기 위해 2012년 매개변수와 2013년 전일 데이터를 이용하여 예측하고자 하는 일단위 토양수분을 구하였다. 그 결과 $R^2=0.646-0.807$, RMSE=1.758-4.802의 정확도를 나타냈다.