Performance Of Time Series Models Research Articles

Effects of missing data imputation methods on univariate blood pressure time series data analysis and forecasting with ARIMA and LSTM

BackgroundMissing observations within the univariate time series are common in real-life and cause analytical problems in the flow of the analysis. Imputation of missing values is an inevitable step in every incomplete univariate time series. Most of the existing studies focus on comparing the distributions of imputed data. There is a gap of knowledge on how different imputation methods for univariate time series affect the forecasting performance of time series models. We evaluated the prediction performance of autoregressive integrated moving average (ARIMA) and long short-term memory (LSTM) network models on imputed time series data using ten different imputation techniques.MethodsMissing values were generated under missing completely at random (MCAR) mechanism at 10%, 15%, 25%, and 35% rates of missingness using complete data of 24-h ambulatory diastolic blood pressure readings. The performance of the mean, Kalman filtering, linear, spline, and Stineman interpolations, exponentially weighted moving average (EWMA), simple moving average (SMA), k-nearest neighborhood (KNN), and last-observation-carried-forward (LOCF) imputation techniques on the time series structure and the prediction performance of the LSTM and ARIMA models were compared on imputed and original data.ResultsAll imputation techniques either increased or decreased the data autocorrelation and with this affected the forecasting performance of the ARIMA and LSTM algorithms. The best imputation technique did not guarantee better predictions obtained on the imputed data. The mean imputation, LOCF, KNN, Stineman, and cubic spline interpolations methods performed better for a small rate of missingness. Interpolation with EWMA and Kalman filtering yielded consistent performances across all scenarios of missingness. Disregarding the imputation methods, the LSTM resulted with a slightly better predictive accuracy among the best performing ARIMA and LSTM models; otherwise, the results varied. In our small sample, ARIMA tended to perform better on data with higher autocorrelation.ConclusionsWe recommend to the researchers that they consider Kalman smoothing techniques, interpolation techniques (linear, spline, and Stineman), moving average techniques (SMA and EWMA) for imputing univariate time series data as they perform well on both data distribution and forecasting with ARIMA and LSTM models. The LSTM slightly outperforms ARIMA models, however, for small samples, ARIMA is simpler and faster to execute.

Phenological Response of an Evergreen Broadleaf Tree, Quercus acuta, to Meteorological Variability: Evaluation of the Performance of Time Series Models

Phenological events are key indicators for the assessment of climate change impacts on ecosystems. Most previous studies have focused on identifying the timing of phenological events, such as flowering, leaf-out, leaf-fall, etc. In this study, we explored the characteristics of the green chromatic coordinate (GCC) values of the evergreen broadleaf tree (Quercus acuta Thunb.), which is a widely used index that serves as a proxy for the seasonal and physiological responses of trees. Additionally, we estimated their relationship with meteorological variables using time series models, including time series decomposition and a seasonal autoregressive integrated moving average with exogenous regressors (SARIMAX). Our results showed that the GCC values and the meteorological variables, which were collected at daily intervals, exhibited a strong autocorrelation and seasonality. This suggests that time series analysis methods are more suitable than ordinary least squares (OLS) regression methods for the fulfillment of statistical assumptions. The time series analysis results highlighted a strong association between precipitation and GCC variation in evergreen broadleaf trees, particularly during the dry season. These results improve our understanding of the response of plant phenology to climate change.

AutoLDT: a lightweight spatio-temporal decoupling transformer framework with AutoML method for time series classification

Time series classification finds widespread applications in civil, industrial, and military fields, while the classification performance of time series models has been improving with the recent development of deep learning. However, the issues of feature extraction effectiveness, model complexity, and model design uncertainty constrain the further development of time series classification. To address the above issues, we propose a Lightweight Spatio-Temporal Decoupling Transformer framework based on Automated Machine Learning technique (AutoLDT). The framework proposes a novel lightweight Transformer with fuzzy position encoding, TS-separable linear self-attention mechanism, and convolutional feedforward network, which mine the temporal and spatial features, as well as the local and global relationship of time series. Fuzzy positional encoding integrates fuzzy ideas to enhance the generalization performance of model information mining. TS-separable linear self-attention mechanism and convolutional feedforward network achieve feature extraction in a lightweight way by decoupling temporal and spatial features of time series. Notably, we adopt the Covariance Matrix Adaptation Evolution Strategy and global adaptive pruning technique to realize automated network structure design, which further improves the model training efficiency and automation, and avoids the uncertainty problem of network design. Finally, we validate the effectiveness of the proposed framework on publicly available UCR and UEA time series datasets. The experimental results show that the proposed framework not only improves the model classification performance in a lightweight way but also dramatically improves the model training efficiency.

Time Series Analysis of Nonlinear Head Dynamics Using Synthetic Data Generated with a Variably Saturated Model.

The performance of time series models is assessed using synthetic head series simulated with a numerical model that solves Richards' equation for variably saturated flow. Heads were simulated in a homogeneous unconfined aquifer between two parallel canals; measured daily precipitation and potential evaporation are specified at the land surface and root water uptake is simulated. The head response to a precipitation event is nonlinear and depends on the saturation degree and rainfall before and after the precipitation event while evaporation reduction occurs during summers. Synthetic series were generated for 27 years and three different soil types; the unsaturated zone thickness varies between 0 and >5 m. The synthetic head series were simulated with a linear and nonlinear time series model. Performance of a linear time series model with four parameters, using a scaled Gamma response, gave R2 values ranging from 0.67 to 0.96. The nonlinear time series model with five parameters simulates recharge using a root zone reservoir after which the head response to recharge is simulated with a scaled Gamma response function. The nonlinear time series model was able to simulate all synthetic head series very well with R2 values above 0.9 for almost all models. The head response of the nonlinear model to a single precipitation event compares well to the response of the variably saturated groundwater model. The provided scripts may be used to simulate synthetic head series for other climates or for systems with additional complexity to assess the performance of other data-driven models.

Forecasting of daily new lumpy skin disease cases in Thailand at different stages of the epidemic using fuzzy logic time series, NNAR, and ARIMA methods.

Lumpy skin disease (LSD) is an important transboundary disease affecting cattle in numerous countries in various continents. In Thailand, LSD is regarded as a serious threat to the cattle industry. Disease forecasting can assist authorities in formulating prevention and control policies. Therefore, the objective of this study was to compare the performance of time series models in forecasting a potential LSD epidemic in Thailand using nationwide data. For the forecasting of daily new cases, fuzzy time series (FTS), neural network auto-regressive (NNAR), and auto-regressive integrated moving average (ARIMA) models were applied to various datasets representing the different stages of the epidemic. Non-overlapping sliding and expanding window approaches were also employed to train the forecasting models. The results showed that the FTS outperformed other models in five of the seven validation datasets based on various error metrics. The predictive performance of the NNAR and ARIMA models was comparable, with NNAR outperforming ARIMA in some datasets and vice versa. Furthermore, the performance of models built from sliding and expanding window techniques was different. This is the first study to compare the forecasting abilities of the FTS, NNAR, and ARIMA models across multiple phases of the LSD epidemic. Livestock authorities and decision-makers may incorporate the forecasting techniques demonstrated herein into the LSD surveillance system to enhance its functionality and utility.

Time Series Prediction of Lung Cancer Death Rates on the Basis of SEER Data.

The purpose of this study was to apply different time series analytical techniques to SEER US lung cancer death rate data to develop a best fit model. Three models for yearly time series predictions were built: autoregressive integrated moving average (ARIMA), simple exponential smoothing (SES), and Holt's double expansional smoothing (HDES) models. The three models were built using Python 3.9, on the basis of Anaconda 2022.10. This study used SEER data from 1975 to 2018 and included 545,486 patients with lung cancer. The best parameters for ARIMA are ARIMA (p, d, q) = (0, 2, 2). In addition, the best parameter for SES was α = .995, whereas the best parameters for HDES were α = .4 and β = .9. The HDES was the model that best fit the lung cancer death rate data, with a root mean square error (RMSE) of 132.91. Including monthly diagnoses, death rates, and years in SEER data increases the number of observations for training and test sets, enhancing the performance of time series models. The reliability of the RMSE was based on the mean lung cancer mortality rate. Owing to the high mean lung cancer death rate of 8,405 patients per year, it is acceptable for reliable models to have large RMSEs.

Cluster-based LSTM models to improve Dengue cases forecast

Public health problems such as dengue fever need accurate forecasts so governments can take effective preventive measures. Deep learning (DL) and machine learning have become increasingly popular as the volume of data increases continuously. Nevertheless, performing accurate predictions in areas with fewer cases can be challenging. When we apply DL models using long short-term memory (LSTM) in different cities considering weekly dengue incidence and climate, some models may present heterogeneous behaviours and poor accuracy because of the need for more data. To mitigate this problem, clustering analysis across time series is performed based on scores to measure the clustering quality in 217 Paraguayan cities. First, we compare the raw and feature-based clustering techniques considering several metrics.Our results indicate that hierarchical clustering combined with Spearman correlation is the most appropriate approach. Finally, several LSTM models built using clustering results were compared. The main contribution of this work is a technique that can improve the performance of time series models that combine information from similar time series and weather data.

A prediction method for the screening current induced field in HTS magnets based on time series models

Due to their special electromagnetic properties, high temperature superconducting (HTS) conductors have become a potential solution for ultra-high field magnet and energy storage applications. However, the screening current induced field (SCIF) has been demonstrated to be the main limitation of high field HTS magnets in actual applications. Based on time series models, this paper presents a prediction method of SCIF to support the design and application of HTS magnets. First, we analyze the data characteristics of the SCIF hysteresis loop. The simulated dataset is prepared for two typical magnet structures: single pancake and solenoid. Then, time series models are proposed for the SCIF prediction. Through intuitive analysis and evaluation metrics, the training performance of time series models is confirmed. After a discussion of hyper-parameters and dimension reduction, the optimized prediction performance is obtained for the SCIF hysteresis loop. In conjunction with the iterative prediction mode, we finally achieve a feasible and effective prediction method of SCIF for HTS magnets. This will provide a tool and research strategy to support the general finite element method.

Machine Learning-Driven Approach for a COVID-19 Warning System

The emergency of the pandemic and the absence of treatment have motivated researchers in all the fields to deal with the pandemic situation. In the field of computer science, major contributions include the development of methods for the diagnosis, detection, and prediction of COVID-19 cases. Since the emergence of information technology, data science and machine learning have become the most widely used techniques to detect, diagnose, and predict the positive cases of COVID-19. This paper presents the prediction of confirmed cases of COVID-19 and its mortality rate and then a COVID-19 warning system is proposed based on the machine learning time series model. We have used the date and country-wise confirmed, detected, recovered, and death cases features for training of the model based on the COVID-19 dataset. Finally, we compared the performance of time series models on the current study dataset, and we observed that PROPHET and Auto-Regressive (AR) models predicted the COVID-19 positive cases with a low error rate. Moreover, death cases are positively correlated with the confirmed detected cases, mainly based on different regions’ populations. The proposed forecasting system, driven by machine learning approaches, will help the health departments of underdeveloped countries to monitor the deaths and confirm detected cases of COVID-19. It will also help make futuristic decisions on testing and developing more health facilities, mostly to avoid spreading diseases.

From zero to hero: Realized partial (co)variances

This paper proposes a generalization of the class of realized semivariance and semicovariance measures introduced by Barndorff-Nielsen et al. (2010) and Bollerslev et al. (2020a) to allow for a finer decomposition of realized (co)variances. The new “realized partial (co)variances” allow for multiple thresholds with various locations, rather than the single fixed threshold of zero used in semi (co)variances. We adopt methods from machine learning to choose the thresholds to maximize the out-of-sample forecast performance of time series models based on realized partial (co)variances. We find that in low dimensional settings it is hard, but not impossible, to improve upon the simple fixed threshold of zero. In large dimensions, however, the zero threshold embedded in realized semi covariances emerges as a robust choice.

Comparing the performance of time series models with or without meteorological factors in predicting incident pulmonary tuberculosis in eastern China

BackgroundMany studies have compared the performance of time series models in predicting pulmonary tuberculosis (PTB), but few have considered the role of meteorological factors in their prediction models. This study aims to explore whether incorporating meteorological factors can improve the performance of time series models in predicting PTB.MethodsWe collected the monthly reported number of PTB cases and records of six meteorological factors in three cities of China from 2005 to 2018. Based on this data, we constructed three time series models, including an autoregressive integrated moving average (ARIMA) model, the ARIMA with exogenous variables (ARIMAX) model, and a recurrent neural network (RNN) model. The ARIMAX and RNN models incorporated meteorological factors, while the ARIMA model did not. The mean absolute percentage error (MAPE) and root mean square error (RMSE) were used to evaluate the performance of the models in predicting PTB cases in 2018.ResultsBoth the cross-correlation analysis and Spearman rank correlation test showed that PTB cases reported in the study areas were related to meteorological factors. The predictive performance of both the ARIMA and RNN models was improved after incorporating meteorological factors. The MAPEs of the ARIMA, ARIMAX, and RNN models were 12.54%, 11.96%, and 12.36% in Xuzhou, 15.57%, 11.16%, and 14.09% in Nantong, and 9.70%, 9.66%, and 12.50% in Wuxi, respectively. The RMSEs of the three models were 36.194, 33.956, and 34.785 in Xuzhou, 34.073, 25.884, and 31.828 in Nantong, and 19.545, 19.026, and 26.019 in Wuxi, respectively.ConclusionsOur study revealed a possible link between PTB and meteorological factors. Taking meteorological factors into consideration increased the accuracy of time series models in predicting PTB, and the ARIMAX model was superior to the ARIMA and RNN models in study settings.

Forecasting the production side of GDP

AbstractWe evaluate the forecasting performance of time series models for the production side of gross domestic product (GDP)—that is, for the sectoral real value‐added series summing up to aggregate output. We focus on two strategies to model a large number of interdependent time series simultaneously: a Bayesian vector autoregressive model (BVAR) and a factor model structure; and compare them to simple aggregate and disaggregate benchmarks. We evaluate point and density forecasts for aggregate GDP and the cross‐sectional distribution of sectoral real value‐added growth in the euro area and Switzerland. We find that the factor model structure outperforms the benchmarks in most tests, and in many cases also the BVAR. An analysis of the covariance matrix of the sectoral forecast errors suggests that the superiority can be traced back to the ability to capture sectoral comovement more accurately.

PERFORMANCE OF KUALA LUMPUR COMPOSITE INDEX STOCK MARKET

Financial Times Stock Exchange (FTSE) Bursa Malaysia Kuala Lumpur Composite Index (KLCI) is made up of over 30 large companies listed on the Bursa Malaysia Main Market. All FTSE Bursa Malaysia data are calculated and disseminated every 15 seconds in real-time. It is believed that the volatility of the stock market has a negative impact on real economic recovery. This paper aims to describe the underlying structure and the phenomenon of the sequence of observations in the series. The information obtained, can determine the performance of time series model to fit the data series from January 2002 until December 2018. Autoregressive Integrated Moving Average (ARIMA) and Generalized Autoregressive Conditional Heteroscedasticity (GARCH) models have been shown to provide the correct trend of volatility. The objectives of this paper are to determine the overall trend of the KLCI stock return and to investigate the performance of Generalized Autoregressive Conditional Heteroscedasticity (GARCH) and Autoregressive Integrated Moving Average (ARIMA) based on KLCI stock return. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) have been chosen to be used in this paper to measure accuracy. The results show that the best ARIMA model is ARIMA(1,1), while for the GARCH model, it is GARCH(1,1).

Exploring the Role of Meteorological Factors in Predicting Incident Pulmonary Tuberculosis: A Time-Series Study in Eastern China

To explore whether incorporating meteorological factors can improve the performance of time series models in predicting pulmonary tuberculosis (PTB), we collected the monthly number of PTB cases and data of six meteorological factors in three cities of China from 2005 to 2018. We constructed three time-series models, including theautoregressive integrated moving average (ARIMA) model, the ARIMA with exogenous variables (ARIMAX) model, and the recurrent neural network (RNN) model. The construction of the ARIMA model did not incorporate meteorological factors, while the construction of ARIMAX and RNN models incorporated meteorological factors. The mean absolute percentage error (MAPE) and root mean square error (RMSE) were used to compare the performance of models in predicting the monthly number of PTB cases in 2018. Both the cross-correlation analysis and spearman rank correlation test showed that PTB was related to meteorological factors in the three cities. Taking meteorological factors into consideration may increase the accuracy of time series models in predicting PTB. Funding Statement: This study was funded by the National Natural Science Foundation of China (81973103), National Key R&D Program of China (2017YFC0907000), Qing Lan Project of Jiangsu Province (2019), and Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). Declaration of Interests: The authors declare no conflict of interest Ethics Approval Statement: This study was approved by the ethics committee of Nanjing Medical University. The study does not include any identifiable personal information. So, there was no requirement for informed consent.

Comparison of forecasting performance of time series models for the wholesale price of dried red peppers: focused on ARX and EGARCH

Comparison of forecasting performance of time series models for the wholesale price of dried red peppers: focused on ARX and EGARCH

Forecasting performance of time series models on electricity spot markets

PurposeStudy conditions of empirical publications on time series modeling and forecasting of electricity prices vary widely, making it difficult to generalize results. The key purpose of the present study is to offer a comparison of different model types and modeling conditions regarding their forecasting performance.Design/methodology/approachThe authors analyze the forecasting performance of AR (autoregressive), MA (moving average), ARMA (autoregressive moving average) and GARCH (generalized autoregressive moving average) models with and without the explanatory variables, that is, power consumption and power generation from wind and solar. Additionally, the authors vary the detailed model specifications (choice of lag-terms) and transformations (using differenced time series or log-prices) of data and, thereby, obtain individual results from various perspectives. All analyses are conducted on rolling calibrating and testing time horizons between 2010 and 2014 on the German/Austrian electricity spot market.FindingsThe main result is that the best forecasts are generated by ARMAX models after spike preprocessing and differencing the data.Originality/valueThe present study extends the existing literature on electricity price forecasting by conducting a comprehensive analysis of the forecasting performance of different time series models under varying market conditions. The results of this study, in general, support the decision-making of electricity spot price modelers or forecasting tools regarding the choice of data transformation, segmentation and the specific model selection.

Improving the Forecast Accuracy of Oil-Exchange Rate Nexus in GCC Countries

This paper renders new evidence on the predictability of GCC dollar exchange rates using crude oil prices relying on the approach of Westerlund [1] [2] that accounts for salient features of the predictor. The results show the presence of significant in-sample predictability of exchange rates using crude oil prices (Brent and WTI prices) across the GCC countries. The results of forecast evaluation based on the root mean square error (RMSE), Campbell-Thompson (C-T) statistic and Diebold-Mariano (D-M) statistic are rather mixed. The superior forecast performance of the oil-based exchange rate model is highly sensitive to the choice of benchmark time-series models. We, however, conclude the overwhelming forecast performance of time-series models (namely, AR, ARMA, and ARFIMA) over our oil-based exchange rate model in predicting exchange rates across the GCC region.

Forecasting performance of time series models on electricity spot markets: a quasi-meta-analysis

PurposeEmpirical publications on the time series modeling and forecasting of electricity prices vary widely regarding the conditions, and the findings make it difficult to generalize results. Against this background, it is surprising that there is a lack of statistics-based literature reviews on the forecasting performance when comparing different models. The purpose of the present study is to fill this gap.Design/methodology/approachThe authors conduct a comprehensive literature analysis from 2000 to 2015, covering 86 empirical studies on the time series modeling and forecasting of electricity spot prices. Various statistics are presented to characterize the empirical literature on electricity spot price modeling, and the forecasting performance of several model types and modifications is analyzed. The key issue of this study is to offer a comparison between different model types and modeling conditions regarding their forecasting performance, which is referred to as a quasi-meta-analysis, i.e. the analysis of analyses to achieve more general findings independent of the circumstances of single studies.FindingsThe authors find evidence that generalized autoregressive conditional heteroscedasticity models outperform their autoregressive–moving-average counterparts and that the consideration of explanatory variables improves forecasts.Originality/valueTo the best knowledge of the authors, this paper is the first to apply the methodology of meta-analyses in a literature review of the empirical forecasting literature on electricity spot markets.

Value-at-Risk Forecasting of Chinese Stock Index and Index Future Under Jumps, Permanent Component, and Asymmetric Information

ABSTRACTThis article investigates the performance of time series models considering the jumps, permanent component of volatility, and asymmetric information in predicting value-at-risk (VaR). We use evaluation statistics including size and variability, accuracy, and efficiency to determine some suitable VaR measures for the Chinese stock index and its futures. The results reveal that models with jumps can provide VaR series that are less average conservative and have higher variability. Furthermore, additional considering the permanent component of volatility and asymmetric effect can induce more accurate and efficient risk measure in the long and short positions of the stock index and its futures.

Semi‐Structural Forecasting of UK Inflation Based on the Hybrid New Keynesian Phillips Curve

AbstractWe develop a semi‐structural model for forecasting inflation in the UK in which the New Keynesian Phillips curve (NKPC) is augmented with a time series model for marginal cost. By combining structural and time series elements we hope to reap the benefits of both approaches, namely the relatively better forecasting performance of time series models in the short run and a theory‐consistent economic interpretation of the forecast coming from the structural model. In our model we consider the hybrid version of the NKPC and use an open‐economy measure of marginal cost. The results suggest that our semi‐structural model performs better than a random‐walk forecast and most of the competing models (conventional time series models and strictly structural models) only in the short run (one quarter ahead) but it is outperformed by some of the competing models at medium and long forecast horizons (four and eight quarters ahead). In addition, the open‐economy specification of our semi‐structural model delivers more accurate forecasts than its closed‐economy alternative at all horizons. Copyright © 2014 John Wiley & Sons, Ltd.