Time Series Models Research Articles

Projection of the prevalence of tracheal, bronchus, and lung cancer incidence using cigarette smoking prevalence in Iran from 1990 to 2018: a comparison of latent period-based models with standard forecasting models

BackgroundSmoking is the major risk factor for tracheal, bronchus, and lung (TBL) cancers. We investigated the feasibility of projecting TBL cancer incidence using smoking incidence rates by incorporating a range of latent periods from the main risk factor exposure to TBL cancer diagnosis.MethodsIn this ecological study, we extracted data on TBL cancer incidence rates in Iran from 1990 to 2018 from the Global Burden of Disease (GBD) database. We also collected data on Iranian cigarette smoking patterns over the past 40 years through a literature review. The weighted average smoking incidence was calculated using a fixed-effects model with Comprehensive Meta-Analysis (CMA) software. Using these data, the five-year TBL cancer incidence in Iran was projected through time series modeling with IT Service Management (ITSM) 2000 software. A second model was developed based on cigarette smoking incidence using linear regression with SPSS (version 22), incorporating different latent periods. The results of these two models were compared to determine the best latent periods.ResultsAn increasing trend in TBL cancer incidence was observed from 2019 to 2023 (first model: 10.30 [95% CI: 9.62, 10.99] to 11.42 [95% CI: 10.85, 11.99] per 100,000 people). In the second model, the most accurate prediction was obtained with latent periods of 17 to 20 years, with the best prediction using a 17-year latent period (10.13 to 11.40 per 100,000 people) and the smallest mean difference of 0.08 (0.84%) per 100,000 people using the standard forecasting model (the ARIMA model).ConclusionProjecting an increase in TBL cancer incidence rates in the future, an optimal latent period of 17 to 20 years between exposure to cigarette smoke and TBL cancer incidence has implications for macrolevel preventive health policymaking to help reduce the burden of TBL cancer in upcoming years.

Complex Network Model of Global Financial Time Series Based on Different Distance Functions

By constructing a complex network model grounded in time series analysis, this study delves into the intricate relationships between the stock markets of 18 countries worldwide. Utilizing 31 distinct time series distance functions to formulate the network, we employ Hamming distance to quantify the resemblance between networks derived from different distance functions. By modulating the network density through distance percentiles (p=0.1, 0.3, 0.5), we demonstrate the similarity of various distance functions across multiple density levels. Our findings reveal that certain distance functions exhibit high degrees of similarity across varying network densities, suggesting their potential for mutual substitution in network construction. Furthermore, the centroid network identified via hierarchical cluster analysis highlights the similarities between the stock markets of different nations, mirroring the intricate interconnections within the global financial landscape. The insights gained from this study offer crucial perspectives for comprehending the intricate network structure of global financial time series data, paving the way for further analysis and prediction of global financial market dynamics.

Forecasting changes in the demographic situation in the Republic of Sakha (Yakutia)

Subject. This article discusses the issues related to the population of the Republic of Sakha (Yakutia). Objectives. The article aims to predict the demographic situation in the Republic of Sakha (Yakutia) on the basis of socio-economic factors. Methods. For the study, we used a statistical analysis, time series regression modeling, and the forecasting method. Results. Based on regression modeling of time series, eleven factors influencing the change in the population of the Republic of Sakha (Yakutia) were identified. The article presents two developed forecasts of changes in the demographic situation. Conclusions. The development of industry, science, and improving the quality of life of the population contribute to the improvement of the demographic situation in the Republic of Sakha (Yakutia). The results obtained can be used to develop a comprehensive development programme for the Republic of Sakha (Yakutia).

Short-Term Power Load Forecasting Method Based on GRU-Transformer Combined Neural Network Model

Load Forecast (LF) is an important task in the planning, control and application of public power systems. Accurate Short Term Load Forecast (STLF) is the premise of safe and economical operation of a power system. In the research of short-term power load forecasting, machine learning and deep learning are the most popular methods at present, but there still exists a problem that the single and simple structure of power load forecasting model leads to low accuracy of load forecasting. In order to improve the accuracy of STLF, a Gated Cycle Unit (GRU)-Transformer combined neural network model is proposed. Transformer encoder structure is used as feature extractor to mine the complex mapping relationships between the input features and load. The advantage of self-attention mechanism is used to solve the problem of information loss of long sequences in short-term power load forecasting. At the same time, the multivariate time series model of GRU is used for model training. The experimental results on the power load data set of a certain region in southwest China and Panama City show that the proposed combined model prediction method has higher accuracy than those proposed in other literatures, which further proves its feasibility and superiority.

Mimicking functional elements of the natural flow regime promotes native fish recovery in a regulated river.

Streamflow regimes that maintain vital functions and processes of aquatic ecosystems are critical to sustaining ecosystem health. In rivers with altered flow regimes, restoring components of the natural flow regime is predicted to conserve freshwater biodiversity by supporting ecological functions and geomorphological processes to which native communities are adapted. However, the effectiveness of environmental flow restoration is poorly understood because of inadequate monitoring and uncertainty in ecological responses to managed changes in specific, quantifiable aspects of the annual streamflow regime. Here, we used time series models to analyze 25 years of fish assemblage data collected before and after environmental flow implementation in a dammed river in California, USA. We examined the response of the fish community to changes in individual components of the flow regime known to support ecosystem functions. We found that as functional flow components shifted toward their predicted natural range, the quasi-extinction risk (likelihood of population declines of >80%) decreased for the native fish assemblage. Following environmental flow implementation, observed changes toward natural ranges of dry season duration, fall pulse flow magnitude, and wet season timing each reduced quasi-extinction risk by at least 40% for the native assemblage. However, functional flow components that shifted away from their predicted natural range, including lower spring recession flows and higher dry season baseflow, resulted in greater quasi-extinction risk for native species. In contrast, non-native species decreased in abundance when flow components shifted toward predicted natural ranges and increased when components shifted away from their natural range. Although most functional flow components remained outside of their natural range following environmental flow implementation, our results indicate that even moderate shifts toward a natural flow regime can benefit native and suppress non-native fish species. Overall, this study provides the most compelling evidence to date of the effectiveness of functional environmental flows in supporting native fish recovery in a highly regulated river.

Impact of the COVID-19 pandemic on antidepressant consumption in the Central region of Portugal: interrupted time series.

To evaluate the impact of the pandemic on the consumption of antidepressive agents in Central Portugal. To estimate the causal effect of the pandemic an interrupted time series analysis was conducted. Data of antidepressant drugs monthly dispensed in community pharmacies between Jan-2010 and Dec-2021 were provided by the regional Health Administration. Anti-Parkinson dopaminergic agents and statins, theoretically not influenced by COVID-19 pandemics, were used as comparator series. The number of packages was converted into defined daily doses and presented as defined daily doses/1000 inhabitants/day. A Bayesian structural time-series model with CausalImpact on R/RStudio was used to predict the counterfactual. Analyses with different geographical granularity (9 sub-regions and 78 municipalities) were performed. When compared to counterfactual, regional consumption non-significantly increased after the pandemic declaration, with a relative effect of + 1.30% [95%CI -1.6%:4.2%]. When increasing the granularity, differences appeared between sub-region with significant increases in Baixo Mondego + 6.5% [1.4%:11.0%], Guarda + 4.4% [1.1%:7.7%] or Cova da Beira + 4.1% [0.17%:8.3%], but non-significant variation in the remaining 6 sub-regions. Differences are more obvious at municipality level, ranging from increases of + 37.00% [32.00%:42.00%] to decreases of -11.00% [-17.00%:-4.20%]. Relative impact positively correlated with percentage of elderly in the municipality (r = 0.301; p = 0.007), and negatively with population density (r=-0.243; p = 0.032). No other predicting variables were found. Antidepressant consumption suffered very slight variations at regional level after the COVID-19 pandemic declaration. Analysis with higher granularity allowed identifying municipalities with higher impact (increase or decrease). The absence of clear association patterns suggests other causal hypotheses of the differences.

Spatio-temporal patterns of excess mortality from non-COVID causes of death in the United States, march-December 2020.

Objectives To estimate the burden of excess mortality from 17 underlying causes of death between March-December 2020 in the United States, and to compare trends in excess deaths from non-COVID causes vs. from COVID-19. Methods Using time series models, we estimated monthly counts of all-cause and cause-specific excess deaths. We stratified by geographic region and compared temporal trends in excess deaths from non-COVID causes to trends in deaths attributed to COVID-19. Results Of approximately 500,000 excess deaths, 70% were attributed to COVID-19. We observed increases in several underlying causes of death, ranging from a 3% increase in kidney disease deaths to a 24% increase in homicides, as well as decreases in deaths from cancer (-0.3%), influenza and pneumonia (-2%), chronic lower respiratory disease (-3%), and suicide (-7%). Trends in excess deaths from cardiovascular disease, diabetes, and Alzheimer's disease closely mirrored trends in deaths from COVID-19. Trends in excess liver disease, homicide, suicide, and motor vehicle accident deaths were negatively correlated with trends in deaths from COVID-19. There was wide regional variation in excess death rates for some causes of death, including a disproportionate increase in homicide and motor vehicle accident deaths in the Great Lakes, and a sustained reduction in cancer deaths in the Mideast and New England. Conclusions Increases in cardiovascular disease, diabetes, and Alzheimer's disease deaths from March-December 2020 likely reflect healthcare system disruptions or acute complications of COVID-19. Excess deaths from homicide and liver disease are more likely to reflect social and economic effects of the emerging pandemic, or other separate causes.

Hurdle GARCH models for nonnegative time series

Hurdle GARCH models for nonnegative time series

The seasonal model of chili price movement with the effect of long memory and exogenous variables for improving time series model accuracy

The seasonal model of chili price movement with the effect of long memory and exogenous variables for improving time series model accuracy

The seasonal model of chili price movement with the effect of long memory and exogenous variables for improving time series model accuracy

This study aimed to explore big-time series data on agricultural commodities with an autocorrelation model comprising long-term processes, seasonality, and the impact of exogenous variables. Among the agricultural commodities with a large amount of data, chili prices exemplified criteria for long-term memory, seasonality, and the impact of various factors on production as an exogenous variable. These factors included the month preceding the new year and the week before the Eid al-Fitr celebration in Indonesia. To address the factors affecting price fluctuations, the Seasonal Autoregressive Fractionally Integrated Moving Average (SARFIMA) model was used to manage seasonality and long-term memory effects in the big data analysis. It improved with the addition of exogenous variables called SARFIMAX (SARFIMA with exogenous variables is known as SARFIMAX). After comparing the accuracy of both models, it was discovered that the SARFIMAX performed better, indicating the influence of seasonality and previous chili prices for an extended period in conjunction with exogenous variables. The SARFIMAX model gives an improvement in model accuracy by adding the effect of exogenous variables. Consequently, this observation concerning price dynamics established the cornerstone for maintaining the sustainability of chili supply even with the big data case.

Efficient Clothing Fashion Prediction using Machine Learning

Fashion trend forecasting is a crucial task for both academia and industry. Although some efforts have been devoted to tackling this challenging task, they only studied limited fashion elements with highly seasonal or simple patterns, which could hardly reveal the real fashion trends. Towards insightful fashion trend forecasting, this work focuses on investigating fine-grained fashion element trends for specific user groups. We first contribute a large-scale fashion trend dataset (FIT) collected from social media with extracted time series fashion element records and user information. Furthermore, to effectively model the time series data of fashion elements with rather complex patterns, we propose a Machine Learning which takes advantage of the capability in modeling time series data. Moreover, it leverages internal and external knowledge in fashion domain that affects the time-series patterns of fashion element trends. Such incorporation of domain knowledge further enhances the deep learning model in capturing the patterns of specific fashion elements and predicting the future trends. Extensive experiments demonstrate that the proposed ML model can effectively capture the complicated patterns of objective fashion elements, therefore making preferable fashion trend forecast.

Evolution of an epidemic: Understanding the opioid epidemic in the United States and the impact of the COVID-19 pandemic on opioid-related mortality.

We conduct this research with a two-fold aim: providing a quantitative analysis of the opioid epidemic in the United States (U.S.), and exploring the impact of the COVID-19 pandemic on opioid-related mortality. The duration and persistence of the opioid epidemic lends itself to the need for an overarching analysis with extensive scope. Additionally, studying the ramifications of these concurrent severe public health crises is vital for informing policies to avoid preventable mortality. Using data from CDC WONDER, we consider opioid-related deaths grouped by Census Region spanning January 1999 to October 2022 inclusive, and later add on a demographic component with gender-stratification. Through the lens of key events in the opioid epidemic, we build an interrupted time series model to reveal statistically significant drivers of opioid-related mortality. We then employ a counterfactual to approximate trends in the absence of COVID-19, and estimate excess opioid-related deaths (defined as observed opioid-related deaths minus projected opioid-related deaths) associated with the pandemic. According to our model, the proliferation of fentanyl contributed to sustained increases in opioid-related death rates across three of the four U.S. census regions, corroborating existing knowledge in the field. Critically, each region has an immediate increase to its opioid-related monthly death rate of at least 0.31 deaths per 100,000 persons at the start of the pandemic, highlighting the nationwide knock-on effects of COVID-19. There are consistent positive deviations from the expected monthly opioid-related death rate and a sizable burden from cumulative excess opioid-related deaths, surpassing 60,000 additional deaths nationally from March 2020 to October 2022, ∼70% of which were male. These results suggest that robust, multi-faceted measures are even more important in light of the COVID-19 pandemic to prevent overdoses and educate users on the risks associated with potent synthetic opioids such as fentanyl.

A Meta-learner approach to multistep-ahead time series prediction

AbstractThe utilization of machine learning has become ubiquitous in addressing contemporary challenges in data science. Moreover, there has been significant interest in democratizing the decision-making process for selecting machine learning algorithms, achieved through the incorporation of meta-features and automated machine learning techniques for both classification and regression tasks. However, this paradigm has not been readily applied to multistep-ahead time series prediction problems. Unlike regression and classification problems, which utilize independent variables not derived from the target variable, time series models typically rely on past values of the series to forecast future outcomes. The structure of a time series is often characterized by features such as trend, seasonality, cyclicality and irregularity. In our study, we illustrate how time series metrics representing these features, in conjunction with an ensemble-based regression Meta-Learner, were employed to predict the standardized mean square error of candidate time series prediction models. Our experiments utilized datasets covering a broad feature space, facilitating the selection of the most effective model by researchers. A rigorous evaluation was conducted to assess the performance of the Meta-Learner on both synthetic and real time series data.

Uncertain statistics models for characterizing annual total energy production

In the field of annual total energy production, it is hard to determine whether the experimental data is linear or nonlinear growth characteristics. Therefore, different regression models are used to test the properties of the observed data. This article uses actual data to perform parameter estimation and residual analysis of linear and nonlinear regression models. In addition, uncertain nonlinear time series model is used to fit the experimental data in order to find a more suitable statistics model after obtaining the growth properties of the observed data. A test is introduced to prove that the disturbance terms presented in the regression and time series models are actually uncertain variables, not random variables. Moreover, the uncertain hypothesis tests show that the uncertain time series model is in better agreement with the real data of the annual total energy production.

Modelling Mixed-Frequency Time Series with Structural Change

AbstractPredictive ability of time series models is easily compromised in the presence of structural breaks, common among financial and economic variables amidst market shocks and policy regime shifts. We address this problem by estimating a semiparametric mixed-frequency model, that incorporate high frequency data either in the conditional mean or the conditional variance equation. The inclusion of high frequency data through non-parametric smoothing functions complements the low frequency data to capture possible non-linear relationships triggered by the structural change. Simulation studies indicate that in the presence of structural change, the varying frequency in the mean model provides improved in-sample fit and superior out-of-sample predictive ability relative to low frequency time series models. These hold across a broad range of simulation settings, such as varying time series lengths, nature of structural break points, and temporal dependencies. We illustrate the relative advantage of the method in predicting stock returns and foreign exchange rates in the case of the Philippines.

Modelling Mixed-Frequency Time Series with Structural Change

AbstractPredictive ability of time series models is easily compromised in the presence of structural breaks, common among financial and economic variables amidst market shocks and policy regime shifts. We address this problem by estimating a semiparametric mixed-frequency model, that incorporate high frequency data either in the conditional mean or the conditional variance equation. The inclusion of high frequency data through non-parametric smoothing functions complements the low frequency data to capture possible non-linear relationships triggered by the structural change. Simulation studies indicate that in the presence of structural change, the varying frequency in the mean model provides improved in-sample fit and superior out-of-sample predictive ability relative to low frequency time series models. These hold across a broad range of simulation settings, such as varying time series lengths, nature of structural break points, and temporal dependencies. We illustrate the relative advantage of the method in predicting stock returns and foreign exchange rates in the case of the Philippines.

Trends and projections of mild and moderate hearing loss among adolescents, young adults, middle-aged adults and age-standardised population in Malaysia from 1996 to 2030.

This study aims to assess the prevalence of mild and moderate hearing loss spanning three decades, from 1990 to 2019, and to project the anticipated trends from 2020 to 2030 among adolescents, young adults, middle-aged adults, and age-standardised groups in Malaysia. This study involved secondary data analysis of mild and moderate hearing loss prevalence over 30 years among the Malaysian population aged 15-19, 25-29, 35-39, 45-49, and age-standardised groups. Subsequently, three time-series models were evaluated and the best models with the minimal Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE) were selected for projecting the prevalence of hearing loss until 2030. A relatively stable trend of mild hearing loss prevalence and gradual decline of moderate hearing loss were observed across all age groups throughout the study period. The prevalence of mild hearing loss was consistently higher than moderate hearing loss across all age groups, with its prevalence increasing with age. The projected prevalence of hearing loss exhibits a gradual declining trend in the future for all age groups, except for mild hearing loss for the 15-19-year-old group. Over the past 30 years, there has been a relatively stable and slightly declining trend in the prevalence of mild and moderate hearing loss among the Malaysian population, respectively with projections showing a slow reduction in the future. These findings highlighted the need for identifying the best intervention and vulnerable age groups, directing increased resources and prioritization towards them.

Unbiased process capability estimation for autocorrelated data using exhaustive systematic sampling

AbstractIt is well known that process control index estimators are inflated when naively applied to positively autocorrelated data. The autocorrelation is a nuisance and not a feature that is captured in the process capability indices. This paper proposes exhaustive systematic sampling to create a pooled variance estimator that replaces the biased estimator of the process data standard deviation when data are autocorrelated. The proposed method is effective because the observations within a systematic sample are spread out in time and should be less correlated with each other as a result. It is similar to Bayesian thinning as a strategy for reducing the impact of autocorrelation except no observations are dropped. Properties of estimated process control indices are derived using quadratic forms and large sample theory that is nonparametric in the sense no distribution or time series model is assumed. Approximately unbiased estimates can be achieved for sufficiently large systematic sampling interval. The proposed method is compared to the time series method in a simulation study that demonstrates similar performance. The proposed method is applied to two examples that use because the target is not the midpoint of the specification limits and the mean differs from the target.

A mixed BAR(1) model driven by serially dependent innovation with application

To better analyze dependent count data with bounded support, this paper proposes a first-order mixture binomial autoregressive (BAR(1)) model, which is constructed via making use of the Pegram operator, where the innovation processes are assumed to be serially dependent to the observation rather than independent. The new model not only inherits the advantages of existing mixture BAR(1) models while also overcoming their shortcomings, but also does well in explaining binomial overdispersion and zero inflation. We infer various statistical properties of the process, including conditional moments, moments and autocovariance functions. The EM algorithm is employed to calculate the conditional maximum likelihood estimate. To evaluate the performance of obtained estimates, some simulation studies are conducted. Furthermore, the forecasting problem for the proposed model is also discussed. To show the applicability of our model, an application to proposition data set is conducted and we compare our model to some competitive integer-valued time series models with bounded support.

Hydraulic system fault diagnosis decoupling method based on 2D time-series modeling and self-attention fusion

Hydraulic systems play a pivotal and extensive role in mechanics and energy. However, the performance of intelligent fault diagnosis models for multiple components is often hindered by the complexity, variability, strong hermeticity, intricate structures, and fault concealment in real-world conditions. This study proposes a new approach for hydraulic fault diagnosis that leverages 2D temporal modeling and attention mechanisms for decoupling compound faults and extracting features from multisample rate sensor data. Initially, to address the issue of oversampling in some high-frequency sensors within the dataset, variable frequency data sampling is employed during the data preprocessing stage to resample redundant data. Subsequently, two-dimensional convolution simultaneously captures both the instantaneous and long-term features of the sensor signals for the coupling signals of hydraulic system sensors. Lastly, to address the challenge of feature fusion with multisample rate sensor data, where direct merging of features through maximum or average pooling might dilute crucial information, a feature fusion and decoupling method based on a probabilistic sparse self-attention mechanism is designed, avoiding the issue of long-tail distribution in multisample rate sensor data. Experimental validation showed that the proposed model can effectively utilize samples to achieve accurate fault decoupling and classification for different components, achieving a diagnostic accuracy exceeding 97% and demonstrating robust performance in hydraulic system fault diagnosis under noise conditions.