Earticle

현재 발생중인 시계열 데이터에 분산변화가 일어날 경우 이동 분산비를 사용하여 분산 변화점을 빠른 시간 내에 탐지하는 문제를 다룬다. 이동 분산비의 분포로서 F분포와 데이터에 의존하여 추정되는 실증적 분포를 제안한 후 상호비교를 통하여, 어느 방법이 시계열 데이터에서 분산의 변화점을 잘 탐지하는지 연구하였다.

This paper considers moving variance ratio (MVR) for valiance detection problem with time series data in progress. For testing purpose, parametric method based on F distribution and nonparametric method based on empirical distribution are compared via simulation study.

It is very important that sellers provide reasonable reserve prices for auction items in internet auction systems. Recently, an agent has been proposed to generate reserve prices automatically based on the case similarity of information retrieval theory and the moving average of time series analysis. However, one problem of the previous approaches is that the recent trend of auction prices is not well reflected on the generated reserve prices, because it simply provides the bid price of the most similar item or an average price of some similar items using the past auction data. In this paper. in order to overcome the problem. we propose a method that generates reserve prices based on the moving average. the exponential smoothing, and the least square of time series analysis. Through performance experiments. we show that the successful bid rate of the new method can be increased by preventing sellers from making unreasonable reserve prices compared with the previous methods.

본 논문에서는 시계열 데이터베이스에서 서브시퀀스 매칭을 효과적으로 처리하는 방안에 관하여 논의한다. 먼저, 본 논문에서는 서브시퀀스 매칭을 위한 기존 기법의 인덱스 검색에서 발생하는 성능상의 문제점들을 지적하고, 이들을 해결할 수 있는 새로운 방법을 제시한다. 제안된 기법은 서브시퀀스 매칭의 인덱스 검색 문제를 윈도우-조인이라는 일종의 공간 조인 문제로 새롭게 해석하는 것에서 출발한다. 윈도우-조인의 빠른 처리를 위하여 제안된 기법에서는 서브시퀀스 매칭을 시작할 때 질의 시퀀스를 위한 R＊-트리를 주기억장치 내에 구성한다. 또한, 제안된 기법은 데이터 시퀀스들을 위한 디스크 상의 R＊-트리와 질의 시퀀스를 위한 주기억장치 상의 R＊-트리를 효과적으로 조인할 수 있는 새로운 알고리즘을 포함한다. 이 알고리즘은 데이터 시퀀스들을 위한 R＊-트리 페이지들을 인덱스 단계의 착오 채택 없이 단 한번만 디스크로부터 액세스하므로 디스크 액세스 측면에서 최적의 기법임이 증명된다. 또한, 다양한 실험을 통한 성능 평가를 통하여 제안된 기법의 우수성을 정량적으로 규명한다.

In this paper, we address efficient processing of subsequence matching in time-series databases. We first point out the performance problems occurring in the index searching of a prior method for subsequence matching. Then, we propose a new method that resolves these problems. Our method starts with viewing the index searching of subsequence matching from a new angle, thereby regarding it as a kind of a spatial-join called a window-join. For speeding up the window-join, our method builds an R＊-tree in main memory for f query sequence at starting of sub-sequence matching. Our method also includes a novel algorithm for joining effectively one R＊-tree in disk, which is for data sequences, and another R＊-tree in main memory, which is for a query sequence. This algorithm accesses each R＊-tree page built on data sequences exactly cure without incurring any index-level false alarms. Therefore, in terms of the number of disk accesses, the proposed algorithm proves to be optimal. Also, performance evaluation through extensive experiments shows the superiority of our method quantitatively.

연관규칙은 데이터 안에 존재하는 항목들간의 종속 관계를 찾아내는 것이다. 기존의 연구에서는 연관규칙 탐사 과정에서 발견항목 자체에만 관심을 두고 연구되어 왔다. 즉, 연관규칙 생성을 위한 후보 항목은 수량을 배제한 항목 대 수량비가 1:1인 상태에서 규칙을 발견하는 연구였다. 이것은 항목의 구매 수량에 관계없이 같은 가중치로 규칙을 발견하는 문제점을 갖고 있다. 두 번째 문제점은 연관규칙은 시간적 연장선상에서 발견되는 규칙이라 할 수 있다. 즉, 규칙을 발견하는 과정에서 모든 자료를 동일한 시간적 가중치를 두어 취급하는 것이다. 본 논문에서는 각각의 아이템을 (아이템, 수량)의 묶음 단위로 후보항목을 만들어 수량적 속성이 포함된 아이템 대 수량 비 1:n의 관계에서 규칙을 발견하는 방법을 제안한다. 또한 과거의 자료들을 이용하여 예측할 때 모든 자료를 동일하게 취급하기보다는 최근의 자료에 더 큰 비중을 주는 예측법을 사용하여 연관규칙 발견의 신뢰성을 높인다. 성능평가는 기존의 알고리즘과 비교하여 제안한 알고리즘의 성능향상 및 타당성을 보인다.

시계열자료가 가진 자기상관은 추정된 상관관계를 왜곡시키는 요인들 중의 하나로 작용한다. 회귀모형의 잔차항에 자기상관이 있는 지를 검정하기 위해 Durbin-Watson 통계량이 흔히 쓰인다. 잔차항에 자기상관을 가진 회귀모형의 효율성을 향상시키기 위해 yule-Walker 법, 비선형최소제곱법, 최우추정법 및 사전백색화법이 사용되어 왔다. 본 연구는 자기상관으로 인한 상관관계의 왜곡을 방지하기 위한 이들 방법들에 대해 고찰하였다. 사전백색화법을 제외한 앞의 3가지 방법을 20년간의 실제 시계열 자료에 적용하였으며 몬테카를로법을 이용하여 각 방법의 오차변이를 조사하였다. 각 방법의 평균잔차제곱분포의 경우, 최우추정법으로 추정된 평균잔차제곱이 가장 작았으며 분포 범위도 가장 작았으나 각 추정방법 사이에 유의한 차이가 발견되지는 않았다.

Autocorrelation in time series data can affect statistical inference in correlation or regression analyses. To improve a regression model from which the residuals are autocorrelated, Yule-Walker method, nonlinear least squares estimation, maximum likelihood method and 'prewhitening' method have been used to estimate the parameters in a regression equation. This study reviewed on the estimation methods of preventing spurious correlation in the presence of autocorrelation and applied the former three methods, Yule-Walker, nonlinear least squares and maximum likelihood method, to a 20-year real data set. Monte carlo simulation was used to compare the three parameter estimation methods. However, the simulation results showed that the mean squared error distributions from the three methods simulated do not differ significantly.

This paper introduces how to adopt the X-12-ARIMA to decompose the railway passenger demand of the Korea National Railroad Especially, selecting on proper filters is focused. The trend filter is identical to the low pass filter in the signal Processing field, and so the seasonal filter is to band pass filter too. Some considerations, selecting a filter, are provided from the view-point of the spectrum analysis. The technique introduced in this paper will be adopted to the project that is to develope the forecasting system of Korea railway passenger demand which is a part of the high speed rail information system.

대용량의 데이터들로부터 사용자가 원하는 데이터를 찾기 위하여 많은 데이터 마이닝 기술들이 연구되어 실제 응용프로그램에서 많이 적용되고 있다. 이러한 데이터 마이닝의 기술 중 연관규칙은 항목들의 집합으로 표현되는 트랜잭션에서 각 항목간의 연관성을 찾는데 사용된다. 그러나 실세계에는 트랜잭션이 없이 일련의 이벤트만 시간에 따라서 발생하는 데이터들이 많이 존재한다. 이러한 시계열 이벤트 데이터들로부터 다양한 가상 트랜잭션을 생성하는 기법들을 제시한다. 이러한 가상 트랜잭션 데이터로 변환된 시계열 데이터에 연관규칙, 순차패턴, 주기패턴과 관련된 여러 가지 알고리즘을 바로 적용 함으로서 유용한 규칙들을 발견해 낼 수 있다.

This paper discusses a data mining technique for time series analysis of traffic data, which provides useful knowledge for network configuration management. Commonly, a network designer must employ a combination of heuristic algorithms and analysis in an interactive manner until satisfactory solutions are obtained. The problem of heuristic algorithms is that it is difficult to deal with large networks and simplification or assumptions have to be made to make them solvable. Various data mining techniques are studied to gain valuable knowledge in large and complex telecommunication networks. In this paper, we propose a traffic pattern association technique among network nodes, which produces association rules of traffic fluctuation patterns among network nodes. Discovered rules can be utilized for improving network topologies and dynamic routing performance.

본 논문은 m개의 독립적인 일차 비선형 시계열로 구성된 패널자료의 동질성 검정에 대한 연구로서 먼저 일반적인 일차 비선형 시계열의 정상성 조건을 유도하고 이어서 동질성 검정법을 제시하고 연관된 극한분포를 규명하였다. 또한 모의실험을 하여 제안된 검정법의 모의검정력을 구하였다.

This evapaper is toluate the forecasting performance of three neural network(NN) approaches against ARIMA model using the famous time series analysis competition data. The first NN approach is to analyze the second Makridakis (M2) Competition Data using Multilayer Perceptron (MLP) that has been the most popular NN model in time series analysis. Since it is recently known that MLP suffers from bias/variance dilemma, two approaches are suggested in this study. The second approach adopts Cascade Correlation Network (CCN) that was suggested by Fahlman & Lebiere as an alternative to MLP. In the third approach, a time series is separated into two series using Noise Filtering Network (NFN) that utilizes autoassociative memory function of neural network. The forecasts in the decomposition analysis are the sum of two prediction values obtained from modeling each decomposed series, respectively. Among the three NN approaches, Decomposition Analysis shows the best forecasting performance on the M2 Competition Data, and is expected to be a promising tool in analyzing socio-economic time series data because it reduces the effect of noise or outliers that is an impediment to modeling the time series generating process.

In this paper, a modified RBF(Radial Basis Function) network structure is suggested for the prediction of a time-series with non-linear, non-stationary characteristics. Coventional RBF network predicting time series by using past outputs sense the trajectory of the time series and react when there exists strong relation between input and hidden activation function's RBF center. But this response is highly sensitive to level and trend of time serieses. In order to overcome such dependencies, hidden activation functions are modified to react to the increments of input variable and multiplied by increment(or dectement) for prediction. When the suggested structure is applied to prediction of Macyey-Glass chaotic time series, Lorenz equation, and Rossler equation, improved performances are obtained.

The objective of this study is to extract information from electroencephalogram(EEG) signals with which we can discriminate mental states. Seven university students were participated in this study. Ten stimuli based on IAPS (International Affective Picture Systems) Were presented at random according to the experimental schedule. 8-channel ($O_1$, $O_2$, $F_3$, $F_4$, $F_7$, $F_8$, $FP_1$, and $FP_2$)EEG signals were recorded at a sampling rate of 204.8 Hz for visual stimuli and analyzed. After random ten sequential stimuli presentation, the subject subjectively assessed the stimulus by scaling from -5 to 5. If the stimulus was the best and the worst, it was scored 5 and -5, respectively. Only maximum and minimum scored-EEG signals within each subject were selected on the basis of subjectively assessment for analysis. EEG signals were transformed into feature objects based on scalar autoregressive model coefficients. They were classified with Discriminant Analysis for each channel. The features produced results with the best classification accuracy of 85.7 % in $O_1$ and $O_2$ for visual stimuli. This study could be extended to establish an algorithm which quantify and classify emotions evoked by visual stimulus using autoregressive models.

In this paper, a modified RBF (Radial Basis Function) neural network structure is suggested for the prediction of time series with non-linear, non-stationary characteristics. Conventional RBF neural network predicting time series by using past outputs is for sensing the trajectory of the time series and for reacting when there exists strong relation between input and hidden neuron's RBF center. But this response is highly sensitive to level and trend of time serieses. In order to overcome such dependencies, hidden neurons are modified to react to the increments of input variable and multiplied by increments(or decrements) of out puts for prediction. When the suggested structure is applied to prediction of Lorenz equation, and Rossler equation, improved performances are obtainable.

경제학에서 분석하는 연도별 국민총생산액, 월별 소비자물가지수, 경영학에서 분석하는 어느 제품의 월별 판매량, 특정주식의 일별 종가 및 거래량, 기상학에서 관찰되는 일별 최고온도 및 최저온도, 태풍의 경로, 등등 여러 학문분야에서 접할 수 있는 통계자료들은 시간이 흐름에 따라 변하는 시계열자료(time series data)들이 많다. 따라서 대부분의 학문분야에서 시계열 분석이 필요하다.(중략)

경제학에서 분석하는 연도별 국민총생산액, 월별 소비자물가지수, 경영학에서 분석하는 어느 제품의 월별 판매량, 특정주식의 일별 종가 및 거래량, 기상학에서 관찰되는 일별 최고온도 및 최저온도, 태풍의 경로, 등등 여러 학문분야에서 접할 수 있는 통계자료들은 시간이 흐름에 따라 변하는 시계열자료(time series data)들이 많다. 따라서 대부분의 학문분야에서 시계열 분석이 필요하다.(중략)

Neural Networks have been advocated as an alternative to statistical forecasting methods. However, the empirical evidences are not consistent. In the present experiments, multi-layered perceptron (MLP) are adopted as approximator to the time series generating processes. To prevent the MLP from being overfitted to the given time series, the information obtained from ARMA modeling is used to determine the architecture of MLP. The proposed approach was tested empirically using the subsamples of the 111 time series used in the first Markridakis Competition. The forecasting results were analyzed to find out the factors that affect the performance of MLP. The experimental results show that the proposed approach outperforms ARMA models in terms of fitting and forecasting accuracy. In addition, it is found that the use of deseasonalized data improves the forecasting accuracy of MLP.

각종 통계패키지 내에 수용되어 있는 시계열 분석방법은 패키지의 특성이나 기능에 따라 다소 차이가 있다. 본 논문에서는 일반덕으로 많이 사용되고 있는 8종류의 통계패키지 (EXECUSTAT, MINITAB, RATS, SAS, SCA, S-PLUS, TSP)에서 시계열 분석이 어떻게 이루어지는지를 비교 검토하였다. 지수평활법과 ARIMA 모형에 의한 분석방법을 중심으로 비교하였으며, 아울러 사용자 관점에서 편리하고 보다 효율적인 패키지가 갖추어야 할 기능들을 제시하였다.

In January 1988, district medical insurance plan was executed on a national scale in Korea. We conducted an evaluation of the impact of execution of district medical insurance plan on number of hospital patients: number of outpatients; and occupancy rate. This study was carried out by Box-Jenkins time series analysis. We tested the statistical significance with intervention component added to ARIMA model. Results of our time series analysis showed that district medical insurance plan had a significant effect on the number of outpatients and occupancy rate. Due to this plan the number of outpatients had increased by 925 patients every month which is equivalent to 8.3 percents of average monthly insurance outpatients in 1987, and occupancy rate had also increased by 0.12 which is equivalent to 16 percents of that in 1987.

To show the distribution and change of blood pressure according to age and to find factors affecting the level of blood pressure in primary school children, a follow-up study was conducted from 1986 to 1989 on 401 first grade children attending primary school in Kangwha County in 1986 and their parents. The blood pressure of the children was significantly increased according to age. The average annual increase was 1.8mmHg in systolic blood pressure and 2.5mmHg in diastolic blood pressure. The level of blood pressure did not show any significant difference in both sexes. Among children who were at or above the 80th percentile of blood pressure in the first grade, 35 and 30% of them have remained at the same level of systolic and diastolic blood pressure respectively in the fourth grade. But we could not find any significance in the tracking of blood pressure of children who were at or above the 90th percentile of blood pressure in the first grade. Weight and pulse rate were shown to be significant factors affecting systolic blood pressure in children of both sexes and mother's blood pressure and skinfold thickness were also affected systolic blood pressure in girls. The variables significantly affecting diastolic blood pressure were arm circumference and pulse rate for boys and height and pulse rate for girls.