• Journal of Intelligence and Information Systems
• /
• v.5 no.1
• /
• pp.103-123
• /
• 1999

Input filtering as a preprocessing method is so much crucial to get good performance in time series forecasting. There are a few preprocessing methods (i.e. ARMA outputs as time domain filters, and Fourier transform or wavelet transform as time-frequency domain filters) for handling time series. Specially, the time-frequency domain filters describe the fractal structure of financial markets better than the time domain filters due to theoretically additional frequency information. Therefore, we, first of all, try to describe and analyze specially some issues on the effectiveness of different filtering methods from viewpoint of the performance of a neural network based forecasting. And then we discuss about neural network model architecture issues, for example, what type of neural network learning architecture is selected for our time series forecasting, and what input size should be applied to a model. In this study an input selection problem is limited to a size selection of the lagged input variables. To solve this problem, we simulate on analyzing and comparing a few neural networks having different model architecture and also use an embedding dimension measure as chaotic time series analysis or nonlinear dynamic analysis to reduce the dimensionality (i.e. the size of time delayed input variables) of the models. Throughout our study, experiments for integration methods of joint time-frequency analysis and neural network techniques are applied to a case study of daily Korean won / U. S dollar exchange returns and finally we suggest an integration framework for future research from our experimental results.

• Communications for Statistical Applications and Methods
• /
• v.25 no.1
• /
• pp.29-42
• /
• 2018

We combine the integer-valued GARCH(1, 1) model with a generalized regime-switching model to propose a dynamic count time series model. Our model adopts Markov-chains with time-varying dependent transition probabilities to model dynamic count time series called the generalized regime-switching integer-valued GARCH(1, 1) (GRS-INGARCH(1, 1)) models. We derive a recursive formula of the conditional probability of the regime in the Markov-chain given the past information, in terms of transition probabilities of the Markov-chain and the Poisson parameters of the INGARCH(1, 1) process. In addition, we also study the forecasting of the Poisson parameter as well as the cumulative impulse response function of the model, which is a measure for the persistence of volatility. A Monte-Carlo simulation is conducted to see the performances of volatility forecasting and behaviors of cumulative impulse response coefficients as well as conditional maximum likelihood estimation; consequently, a real data application is given.

• Convergence Security Journal
• /
• v.11 no.3
• /
• pp.19-24
• /
• 2011

Software failure time presented in the literature exhibit either constant monotonic increasing or monotonic decreasing, For data analysis of software reliability model, data scale tools of trend analysis are developed. The methods of trend analysis are arithmetic mean test and Laplace trend test. Trend analysis only offer information of outline content. In this paper, we discuss forecasting failure time case of failure time censoring. In this study, time series analys is used in the simple moving average and weighted moving averages, exponential smoothing method for predict the future failure times, Empirical analysis used interval failure time for the prediction of this model. Model selection using the mean square error was presented for effective comparison.

• The Korean Journal of Applied Statistics
• /
• v.9 no.1
• /
• pp.125-137
• /
• 1996

Time series data are often contaminated with outliers due to influence of unusal and non-responsitive events. The effect of the outliers is larger in the time series analysis than in the other statistical analysis, because the time series data have dependent structure over time. This paper focuses on the effect of patchy outliers on forecasting. Especially, the increase of the mean square of the l-step-ahead forecast error is derived and used to evaluate the impact of those outliers on the forecast. We fine, in general, that this increase is rather small, provided that the patchy outliers does not occur too close to the forecast origin.

• Journal of the Korea Convergence Society
• /
• v.6 no.4
• /
• pp.161-167
• /
• 2015

It is very difficult to predict time-series data. This is because data obtained from the signal having a non-linear characteristic has an uncertainty. In this paper, By differentiating time-series data is the average of the past data under the premise that change depending on what pattern, and find the soft look of time-series change pattern. This paper also apply the probability variables to generalize time-series data having a specific data according to the reflection ratio of the differentiation. The predicted value is estimated by removing cyclic movement and seasonal fluctuation, and reflect the trend by extracting the irregular fluctuation. Predicted value has demonstrated the superiority of the proposed algorithm and compared with the best results by a simple moving average and the moving average.

• Smart Structures and Systems
• /
• v.21 no.5
• /
• pp.705-713
• /
• 2018

High-speed rail (HSR) has been in operation and development in many countries worldwide. The explosive growth of HSR has posed great challenges for operation safety and ride comfort. Among various technological demands on high-speed trains, vibration is an inevitable problem caused by rail/wheel imperfections, vehicle dynamics, and aerodynamic instability. Ride comfort is a key factor in evaluating the operational performance of high-speed trains. In this study, online monitoring data have been acquired from an in-service high-speed train for condition assessment. The measured dynamic response signals at the floor level of a train cabin are processed by the Sperling operator, in which the ride comfort index sequence is used to identify the train's operation condition. In addition, a novel technique that incorporates salient features of Bayesian inference and time series analysis is proposed for outlier detection and change detection. The Bayesian forecasting approach enables the prediction of conditional probabilities. By integrating the Bayesian forecasting approach with time series analysis, one-step forecasting probability density functions (PDFs) can be obtained before proceeding to the next observation. The change detection is conducted by comparing the current model and the alternative model (whose mean value is shifted by a prescribed offset) to determine which one can well fit the actual observation. When the comparison results indicate that the alternative model performs better, then a potential change is detected. If the current observation is a potential outlier or change, Bayes factor and cumulative Bayes factor are derived for further identification. A significant change, if identified, implies that there is a great alteration in the train operation performance due to defects. In this study, two illustrative cases are provided to demonstrate the performance of the proposed method for condition assessment of high-speed trains.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.4
• /
• pp.685-694
• /
• 2009

The idea of this paper is to get the time series data from the number of student on the elementary, meddle and high-school for the forecasting of the numbers of student. Tow models, model A and model B, of time series data are obtained. The Holt-Winters additive methods are used for the forecasting of the numbers of student with the model A and model B until 2019 year. As the result, the abilities of forecasting on model A and B are better than those of the Korean education statistical system 2007.

• Journal of Industrial Technology
• /
• v.28 no.B
• /
• pp.101-109
• /
• 2008

In general, non-stationary or chaos time series forecasting is very difficult since there exists a drift and/or nonlinearities in them. To overcome this situation, we suggest a new prediction method based on multiple model TS fuzzy predictors combined with preprocessing of time series data, where, instead of time series data, the differences of them are applied to predictors as input. In preprocessing procedure, the candidates of optimal difference interval are determined by using con-elation analysis and corresponding difference data are generated. And then, for each of them, TS fuzzy predictor is constructed by using k-means clustering algorithm and least squares method. Finally, the best predictor which minimizes the performance index is selected and it works on hereafter for prediction. Computer simulation is performed to show the effectiveness and usefulness of our method.

• KDI Journal of Economic Policy
• /
• v.14 no.1
• /
• pp.121-145
• /
• 1992

This paper examines characteristics of time series data related to the construction investment(stationarity and time series components such as secular trend, cyclical fluctuation, seasonal variation, and random change) and surveys predictibility, fitness, and explicability of independent variables of various models to build a short-term construction investment forecasting model suitable for current economic circumstances. Unit root test, autocorrelation coefficient and spectral density function analysis show that related time series data do not have unit roots, fluctuate cyclically, and are largely explicated by lagged variables. Moreover it is very important for the short-term construction investment forecasting to grasp time lag relation between construction investment series and leading indicators such as building construction permits and value of construction orders received. In chapter 3, we explicate 7 forecasting models; Univariate time series model (ARIMA and multiplicative linear trend model), multivariate time series model using leading indicators (1st order autoregressive model, vector autoregressive model and error correction model) and multivariate time series model using National Accounts data (simple reduced form model disconnected from simultaneous macroeconomic model and VAR model). These models are examined by 4 statistical tools that are average absolute error, root mean square error, adjusted coefficient of determination, and Durbin-Watson statistic. This analysis proves two facts. First, multivariate models are more suitable than univariate models in the point that forecasting error of multivariate models tend to decrease in contrast to the case of latter. Second, VAR model is superior than any other multivariate models; average absolute prediction error and root mean square error of VAR model are quitely low and adjusted coefficient of determination is higher. This conclusion is reasonable when we consider current construction investment has sustained overheating growth more than secular trend.

• Korea Trade Review
• /
• v.46 no.6
• /
• pp.191-209
• /
• 2021

In this paper, various time series analysis models and machine learning models are presented for long-term prediction of export growth rate, and the prediction performance is compared and reviewed by RMSE and MAE. Export growth rate is one of the major economic indicators to evaluate the economic status. And It is also used to predict economic forecast. The export growth rate may have a negative (-) value as well as a positive (+) value. Therefore, Instead of using the ReLU function, which is often used for time series prediction of deep learning models, the PReLU function, which can have a negative (-) value as an output value, was used as the activation function of deep learning models. The time series prediction performance of each model for three types of data was compared and reviewed. The forecast data of long-term prediction of export growth rate was deduced by three forecast methods such as a fixed forecast method, a recursive forecast method and a rolling forecast method. As a result of the forecast, the traditional time series analysis model, ARDL, showed excellent performance, but as the time period of learning data increases, the performance of machine learning models including LSTM was relatively improved.