Class Of Time Series Models Research Articles

A Novel Stochastic Fuzzy Time Series Forecasting Model Based on a New Partition Method

Fuzzy Time Series (FTS) models are commonly used in time series forecasting, where they do not require any statistical assumptions on time series data. FTS models can handle data sets with a small number of observations or with uncertainty. This is a general advantage of FTS as compared with other techniques. However, FTS models still have some criticisms, such as the optimal lengths of intervals and the proper weights, which always influence the model accuracy and still have been of many concerns in literature. The work in this paper proposes a novel FTS forecasting model based on a new tree partitioning method (TPM) and Markov chain (MC), called FTSMC-TPM, for determining the optimal partitions of intervals and the proper weights vectors respectively, and this will greatly improve the model accuracy. The efficiency of the FTSMC-TPM model is tested using two types of time series consisting of the air pollution index (API) data, which is collected from Kuala Lumpur, Malaysia and the benchmark data of the yearly enrollments for the University of Alabama. Three statistical criteria have been used for investigating the accuracy of the proposed model. The results indicate that the proposed model outperforms the existing classic and advanced time series models in terms of forecasting accuracy. In addition, the proposed model shows the ability to successfully deal with forecasting problems to obtain higher model accuracy, which is examined in comparison with the existing models to validate its superiority. Hence, this study demonstrates that the proposed model is more suitable for the accurate prediction of air pollution events as well as for forecasting any type of random time series.

Forecasting house prices in Turkey: GLM, VAR and time series approaches

Purpose- A wide range of decision-makers is interested in educated forecasts for house prices. The technical analysis introduced in this study aims to estimate future (forecasted) house prices and provide sufficient evidence in support of the adequacy of the estimated models obtained from parametric and non-parametric modeling methods for Turkey's housing market. Methodology- We employ non-parametric and various time series methods to find appropriate fits to forecast Turkey's house price index (HPI). In our modelling, we consider macroeconomic indicators related to housing markets, such as; gold, interest rate and currencies. In this study, first using the explanatory variables, we construct two Generalized Linear Models (GLM) and a Vector Auto Regressive (VAR) model. Then, we construct two univariate time series models. HPI series inherits seasonality. Even though the HPI contains seasonality, first, we neglect the seasonal effect and come up an Autoregressive Moving Average (ARMA(p,q)) model among many other alternative ARMA models. Second, we consider the seasonality effect on the housing market index and construct a seasonal Autoregressive Integrated Moving Average (ARIMA(p,d,q)(P,D,Q)) and exponential smoothing models. Findings- The analysis identifies forecasts of Turkey’s housing market index from both the seasonal ARIMA(p,d,q) (P,D,Q)_m and Holt Winter models as accurate models compared to classical time series models, namely ARIMA(p,d,q) models, based on the explanation power measure (R^2) values and out-of-sample error measures MSE, RMSE and MAE. Conclusion- The study has three main contributions: i) Our forecast shows Turkey's housing market's return will not increase in the following 12-months. ii) The seasonal ARIMA and exponential smoothing models forecast some negative returns within the given forecasting period, which should be considered a warning for Turkey's housing market for the future. iii) GLM and VAR models illustrate that Turkey's housing market shows a high dependence on gold, inflation, and foreign exchange rates than other well-known economic indicators.

Classification and Forecasting Model for Covid -19 Disease Severities based on Medical Diagnosis using Weighted Average Dynamic Time Warping Technique

Due to Covid-19 pandemic, mankind was affected with respect to mental and physical stress. The causes of the disease have to be analysed based on the correlation factors such as Medication, Age, Gender, physical fitness and habits. In this paper we have proposed a classification model based on weighted average dynamic time warping approach to detect the disease severity as High, Medium, and Low by considering the multi-variant dependent variables that affect the prediction of Covid-19 positive cases. We also proposed the forecasting model based on the time series exponential moving average to identify the growth of disease with respect to Age, Gender and Medical history of the covid-19 positive patient. The results are obtained by defining the correlation function to measure the disease severity in the range of High, Medium and Low. The time series analysis is done with respect to mean average disease severity and also number of positive cases. The forecasting is performed based on the age, gender and existing disorders in health. The results are analysed with other time series classification models such as weighted time wrapping to make the model fits maximum to the available input.

Time Series Forecasting and Classification Models Based on Recurrent with Attention Mechanism and Generative Adversarial Networks.

Time series classification and forecasting have long been studied with the traditional statistical methods. Recently, deep learning achieved remarkable successes in areas such as image, text, video, audio processing, etc. However, research studies conducted with deep neural networks in these fields are not abundant. Therefore, in this paper, we aim to propose and evaluate several state-of-the-art neural network models in these fields. We first review the basics of representative models, namely long short-term memory and its variants, the temporal convolutional network and the generative adversarial network. Then, long short-term memory with autoencoder and attention-based models, the temporal convolutional network and the generative adversarial model are proposed and applied to time series classification and forecasting. Gaussian sliding window weights are proposed to speed the training process up. Finally, the performances of the proposed methods are assessed using five optimizers and loss functions with the public benchmark datasets, and comparisons between the proposed temporal convolutional network and several classical models are conducted. Experiments show the proposed models’ effectiveness and confirm that the temporal convolutional network is superior to long short-term memory models in sequence modeling. We conclude that the proposed temporal convolutional network reduces time consumption to around 80% compared to others while retaining the same accuracy. The unstable training process for generative adversarial network is circumvented by tuning hyperparameters and carefully choosing the appropriate optimizer of “Adam”. The proposed generative adversarial network also achieves comparable forecasting accuracy with traditional methods.

Short-Term Firm-Level Energy-Consumption Forecasting for Energy-Intensive Manufacturing: A Comparison of Machine Learning and Deep Learning Models

To minimise environmental impact, to avoid regulatory penalties, and to improve competitiveness, energy-intensive manufacturing firms require accurate forecasts of their energy consumption so that precautionary and mitigation measures can be taken. Deep learning is widely touted as a superior analytical technique to traditional artificial neural networks, machine learning, and other classical time-series models due to its high dimensionality and problem-solving capabilities. Despite this, research on its application in demand-side energy forecasting is limited. We compare two benchmarks (Autoregressive Integrated Moving Average (ARIMA) and an existing manual technique used at the case site) against three deep-learning models (simple Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)) and two machine-learning models (Support Vector Regression (SVR) and Random Forest) for short-term load forecasting (STLF) using data from a Brazilian thermoplastic resin manufacturing plant. We use the grid search method to identify the best configurations for each model and then use Diebold–Mariano testing to confirm the results. The results suggests that the legacy approach used at the case site is the worst performing and that the GRU model outperformed all other models tested.

Air Transportation Direct Share Time Series Forecasting: A Hybrid Model

In modern air transportation, the direct share is the ratio of direct passengers to total passengers on a directional origin and destination (O&D) pair. The forecasting of direct share time series on the O&D level, as part of the detailed demand forecasting, plays a fundamental role in air transportation planning and development. An accurate forecasting of the O&D direct share time series can benefit the air transportation planners, airlines, and airports in multiple ways. Based on the previous analysis, the direct share time series is O&D specific. This research focuses on developing accurate direct share time series forecasting models on O&D markets with different characteristics. Both classical time series models and supervised learning regression models are investigated carefully. A novel hybrid model that combines time series concept and machine learning modeling techniques is proposed, which can provide more accurate forecasting performance and valuable insights into the O&D markets. To automatically select the forecasting model for each O&D pair, a general modeling framework is proposed for direct share time series forecasting. Based on the forecasting performance comparison, the modeling framework can provide promising direct share time series forecasting, which is a reliable replacement for the model used in the Federal Aviation Administration Terminal Area Forecast.

Analysis and prediction of confirmed COVID-19 cases in China with uncertain time series

This paper presents an uncertain time series model to analyse and predict the evolution of confirmed COVID-19 cases in China, excluding imported cases. Compared with the results of the classical time series model, the uncertain time series model could better describe the COVID-19 epidemic by using an uncertain hypothesis test to filter out outliers. This improvement is reflected in the two observations. One is that the estimated variance of the disturbance term in the uncertain time series model is more appropriate and acceptable than that in the classical time series model, and the other is that the disturbance term of the classical time series model cannot be regarded as a random variable but as an uncertain variable.

Large-scale multivariate forecasting models for Dengue - LSTM versus random forest regression

Effective management of seasonal diseases such as dengue fever depends on timely deployment of control measures prior to the high transmission season. As the epidemic season fluctuates from year to year, the availability of accurate forecasts of incidence can be decisive in attaining control of such diseases. Obtaining such forecasts from classical time series models has proven a difficult task. Here we propose and compare machine learning models incorporating feature selection,such as LASSO and Random Forest regression with LSTM a deep recurrent neural network, to forecast weekly dengue incidence in 790 cities in Brazil. We use multivariate time-series as predictors and also utilize time series from similar cities to capture the spatial component of disease transmission. The LSTM recurrent neural network model attained the highest performance in predicting future incidence on dengue in cities of different sizes.

Goodness-of-fit testing for time series models via distance covariance

In many statistical modeling frameworks, goodness-of-fit tests are typically administered to the estimated residuals. In the time series setting, whiteness of the residuals is assessed using the sample autocorrelation function. For many time series models, especially those used for financial time series, the key assumption on the residuals is that they are in fact independent and not just uncorrelated. In this paper, we apply the auto-distance covariance function (ADCV) to evaluate the serial dependence of the estimated residuals. Distance covariance can discriminate between dependence and independence of two random vectors. The limit behavior of the test statistic based on the ADCV is derived for a general class of time series models. One of the key aspects in this theory is adjusting for the dependence that arises due to parameter estimation. This adjustment has essentially the same form regardless of the model specification. We illustrate the results in simulated examples.

Variance Homogeneity Test Based on Cumulative Wavelet Coefficients

The aim of this paper is to address the problem of variance break detection in time series in wavelet domain. The maximal overlapped discrete wavelet transform (MODWT) decomposes the series variance across scales into components known as the wavelet variances. We introduce all scale wavelet coefficients based test statistic that allows detecting a break in the homogeneity of the variance of a series through changes in the mean of wavelet variances. The statistic makes use of the traditional CUSUM (cumulative sum) based test designed to test for a break in the mean and constructed using cumulative sums of the square of wavelet coefficients. Under moments and mixing conditions, the test statistic satisfies the functional central limit theorem (FCLT) for a broad class of time series models. The overall performance of our test statistic is compared to the traditional Inclan [8] test statistic. The effectiveness of our statistic is supported by good performances reported in simulations and is as reliable as the traditional statistic. Our method provides a nonparametric test procedure that can be applied to a large class of linear and non linear models. We illustrate the practical use of our test procedure with the quarterly percentage changes in the Americans personal savings data set over the period 1970-2016. Both statistics detect a break in the variance in the second quarter of 2001.

Markov Weighted Fuzzy Time-Series Model Based on an Optimum Partition Method for Forecasting Air Pollution

Air pollution is one of the main environmental issues faced by most countries around the world. Forecasting air pollution occurrences is an essential topic in air quality research due to the increase in awareness of its association with public health effects, and its development is vital to managing air quality. However, most previous studies have focused on enhancing accuracy, while very few have addressed uncertainty analysis, which may lead to insufficient results. The fuzzy time-series model is a better option in air pollution forecasting. Nevertheless, it has a limitation caused by utilizing a random partitioning of the universe of discourse. This study proposes a novel Markov weighted fuzzy time-series model based on the optimum partition method. Fitting the optimum partition method has been done based on five different partition methods via two stages. The proposed model is first applied for forecasting air pollution using air pollution index (API) data collected from an air monitoring station located in Klang city, Malaysia. The performance of the proposed model is evaluated based on three statistical criteria, which are the mean absolute percentage error, mean squared error and Theil’s U statistic, using the daily API data. For further validation of the model, it is also implemented for benchmark enrolment data from the University of Alabama. According to the analysis results, the proposed model greatly improved the performance of air pollution index and enrolment prediction accuracy, for which it outperformed several state-of-the-art fuzzy time-series models and classic time-series models. Thus, the proposed model could be a better option for air quality forecasting for managing air pollution.

Drought modeling using classic time series and hybrid wavelet-gene expression programming models

The standardized precipitation evapotranspiration index (SPEI) at three different time scales (i.e., SPEI-3, SPEI-6, and SPEI-12) from six meteorology stations located in Turkey are modeled in this study. To this end, two types of classic time series models, namely linear autoregressive (AR) and non-linear bi-linear (BL) are used. Furthermore, the hybrid models are proposed by coupling the wavelet (W) and gene expression programming (GEP). Five various mother wavelets (i.e., Haar, db4, Symlet, Meyer, and Coifflet), for the first time, are employed and compared for implementing the hybrid W-GEP approach in drought modeling. The modeling results of SPEI droughts via the time series models illustrated that the non-linear BL performs slightly better than the linear AR. Moreover, all the hybrid W-GEP models developed in the study region provide superior performances compared to the standalone GEP. In general, db4 in SPEI-3 modeling and Symlet for modeling the SPEI-6 and SPEI-12 of the studied locations are the optimal wavelets to develop the W-GEP. Finally, the SPEI series at each target station is modeled through applying the SPEI data of the five neighboring stations. It is found that the SPEI data of neighboring stations are appropriate for modeling the SPEI series of the target station when the SPEI data of the target station is not at hand. For this case, the performance of standalone GEP for modeling the SPEI-3 and SPEI-6 of the stations is generally better than the case of utilizing the original SPEI data at each target station.

Adversarial Attacks on Time Series.

Time series classification models have been garnering significant importance in the research community. However, not much research has been done on generating adversarial samples for these models. These adversarial samples can become a security concern. In this paper, we propose utilizing an adversarial transformation network (ATN) on a distilled model to attack various time series classification models. The proposed attack on the classification model utilizes a distilled model as a surrogate that mimics the behavior of the attacked classical time series classification models. Our proposed methodology is applied onto 1-nearest neighbor dynamic time warping (1-NN DTW) and a fully convolutional network (FCN), all of which are trained on 42 University of California Riverside (UCR) datasets. In this paper, we show both models were susceptible to attacks on all 42 datasets. When compared to Fast Gradient Sign Method, the proposed attack generates a larger faction of successful adversarial black-box attacks. A simple defense mechanism is successfully devised to reduce the fraction of successful adversarial samples. Finally, we recommend future researchers that develop time series classification models to incorporating adversarial data samples into their training data sets to improve resilience on adversarial samples.

Time-series machine-learning error models for approximate solutions to parameterized dynamical systems

This work proposes a machine-learning framework for modeling the error incurred by approximate solutions to parameterized dynamical systems. In particular, we extend the machine-learning error models (MLEM) framework proposed in Ref. Freno and Carlberg (2019) to dynamical systems. The proposed Time-Series Machine-Learning Error Modeling (T-MLEM) method constructs a regression model that maps features – which comprise error indicators that are derived from standard a posteriori error-quantification techniques – to a random variable for the approximate-solution error at each time instance. The proposed framework considers a wide range of candidate features, regression methods, and additive noise models. We consider primarily recursive regression techniques developed for time-series modeling, including both classical time-series models (e.g., autoregressive models) and recurrent neural networks (RNNs), but also analyze standard non-recursive regression techniques (e.g., feed-forward neural networks) for comparative purposes. Numerical experiments conducted on multiple benchmark problems illustrate that the long short-term memory (LSTM) neural network, which is a type of RNN, outperforms other methods and yields substantial improvements in error predictions over traditional approaches.

Benchmarking Machine Learning Algorithms on Blood Glucose Prediction for Type I Diabetes in Comparison With Classical Time-Series Models.

This paper aims to compare the performance of several commonly known machine-learning (ML) models versus a classic Autoregression with Exogenous inputs (ARX) model in the prediction of blood glucose (BG) levels using time-series data of patients with Type 1 diabetes (T1D). The ML algorithms include ML-based regression models and deep learning models such as a vanilla Long-Short-Term-Memory (LSTM) Network and a Temporal Convolution Network (TCN). Evaluations have been conducted with respect to different input features, regression model orders, as well as using the recursive method or direct method for multi-step prediction of BG levels. Prediction performance metrics include the average Root Mean Square Error (RMSE), temporal gain (TG) for early prediction, and the normalized energy of the second-order differences (ESOD) of the predicted time series to reflect risk of false alerts on hypo/hyper glycemia events. The ARX model achieved the lowest average RMSE for both recursive and direct methods, the second highest average TG under the direct method, but with a higher average normalized ESOD than some other models. There was no significant advantage observed from the ML models compared to the classic ARX model in predicting BG levels for T1D, except that TCN's performance was more robust with respect to BG trajectories with spurious oscillations, for which ARX tended to over-predict peak BG values and under-predict valley BG values. Insight learned from this study could help researchers and clinical practitioners to select appropriate models for BG prediction.

Comparing classic time series models and the LSTM recurrent neural network: An application to S&P 500 stocks

In the financial literature, there is great interest in the prediction of stock prices. Stock prediction is necessary for the creation of different investment strategies, both speculative and hedging ones. The application of neural networks has involved a change in the creation of predictive models. In this paper, we analyze the capacity of recurrent neural networks, in particular the long short-term recurrent neural network (LSTM) as opposed to classic time series models such as the Exponential Smooth Time Series (ETS) and the Arima model (ARIMA). These models have been estimated for 284 stocks from the S&P 500 stock market index, comparing the MAE obtained from their predictions. The results obtained confirm a significant reduction in prediction errors when LSTM is applied. These results are consistent with other similar studies applied to stocks included in other stock market indices, as well as other financial assets such as exchange rates.

Comparitive Study of Time Series and Deep Learning Algorithms for Stock Price Prediction

Stock Price Prediction has always been an intriguing research problem in financial domain. In the past decade, various methodologies based on classical time series, machine learning, deep learning and hybrid models which constitute the combinations of algorithms have been proposed with reasonable effectiveness in predicting the stock price. There is also considerable research work in comparing the performances of these models. However, from literature review, stems a concern, that is, lack of formal methodology that allows comparison of performances of the different models. For example, the lack of guidance on the generalizability of the time series models and optimised deep learning models is concerning. In addition, there is also a lack of guidance on general fitment of models, which can vary in accordance with forecasting requirement of stock price. This study is aimed at establishing a formal methodology of comparing different types of time series forecasting models based on like for like paradigm. The effectiveness of Deep Learning and Time-Series models have been evaluated by predicting the close prices of three banking stocks. The characteristics of the models in terms of generalizability are compared. The impact of the forecasting period on performance for various models are evaluated on a common metric. In most of the previous studies, the forecasting was done for the periods of 1 day, 5 days or 31 days. To keep the impact of volatility in the stock market due to various political and economic shocks both at international and domestic domains to the minimum, the forecasting periods of up 2 days for short term and 5 days for long term are considered. It has been evidenced that the deep learning models have outperformed time series models in terms of generalisability as well as short- and long-term forecasts.

The Ukraine and EU trade balance: prediction via various models of time series

This article is the first to study, simulate and forecast the monthly dynamics of the trade balance between Ukraine and the European Union for the period from 2005 to 2019. In the presented work, three types of models were used for modeling and forecasting: Automated Neural Networks, additive models ARIMA *ARIMAS (Autoregressive integrated moving average with season component) and Holts model with a damped trend. When modeling using the Automated Neural Networks module, various ensembles of networks and neuron activation functions in hidden layers were used. It turned out that Automated Neural Networks have poor prognostic ability (as in the case considered by us, when modeling insufficiently long series of dynamics). Therefore, when modeling and forecasting the dynamics of the Ukraine-EU trade balance, classical (so-called Box-Jenkins) time series models were used. In this case, the time series is divided into several components (in our case, terms): the main trend is the trend, the seasonal component and the random component (the so-called white noise). By smoothing the initial series, a trend was found, and an analysis of the autocorrelation functions revealed a one-year seasonality. Eliminating the trend and the seasonal component, we obtained a random component, which has a Gaussian distribution. This made it possible to apply first the ARIMA* ARIMAS additive model, and then the Holt model of exponential smoothing with a damped trend. Adequate models of Ukraine-EU trade balance dynamics have been obtained, according to which the forecast has been made. A comparative analysis of the models used. The best model was chosen for forecasting, which allowed to get a good forecast (in comparison with actual data).

To Improve Voice Recognition System using GMM and HMM Classification Models

In this paper, the researcher study automatic speech recognition technology for the individual. We propose a new voice recognition system using a hybrid model GMM-HMM. HMM and GMM is a non-linear classification model. Each state in an HMM can be thought of as a GMM. HMM is consider observation for state. It is also known as time series classification model. In this model, samples have been trained independently and parameters consider jointly which provides better performance than other classification models. Speech recognition system consider two types of learning patterns such as supervised learning and unsupervised learning. In this context speaker dependent and speaker independent used for identifying the efficient and effective voice. In this paper researcher considered supervised learning model for recognize efficient voice. This new voice recognition system identifies incorrect phonemes and verifies the correctness of voice pronunciation. Using the GMM-HMM hybrid model produces better performance and effectiveness of voice.

Dynamic conditional score models: a review of their applications

ABSTRACTIn this paper, applications of dynamic conditional score (DCS) models are reviewed and those models are discussed in relation to classical time series models from the literature. DCS models are robust to outliers, which improves their statistical performance compared to classical models. Three applications are presented in order to compare the statistical performances of DCS and classical models in three very different contexts: (i) The QAR (quasi-autoregressive) plus Beta-t-EGARCH (exponential autoregressive conditional heteroscedasticity) model is presented, which is a score-driven expected return plus volatility model. This model is used for daily returns on the DAX (Deutscher Aktienindex) equity index for the period of January 1988 to December 2017. (ii) The score-driven local level and seasonality plus Beta-t-EGARCH model is presented, which is used for daily AFN/USD (Afghan Afghani/United States Dollar) currency exchange rates for the period of March 2007 to July 2017. (iii) The Seasonal-t-QVAR (quasi-vector autoregressive) model is presented, which is a score-driven multivariate dynamic model of location. For this model, monthly US inflation rate and US unemployment rate are used for the period of January 1948 to December 2017. For all applications, the statistical performance of each DCS model is superior to that of a corresponding classical alternative.