The approximation of an isolated epidemic process wave using a combination of exponents

The most commonly used methods for the medium- and long-term forecasting of epidemic processes are based on the classical SIR (susceptible – infected – recovered) model and its numerous modifications. In this approach, the dynamics of the epidemic is approximated using the solutions of differential or discrete equations. The forecasting methods based on the approximation of data by functions of a given class are usually focused on obtaining a short-term forecast. They are not used for the long-term forecasts of epidemic processes due to their insufficient efficiency for forecasting nonstationary processes. In this paper, we formulated a hypothesis that the primary waves of the COVID-19 pandemic, which took place in a number of European countries, including the Republic of Belarus, in the spring-summer of 2020 are isolated and therefore can be regarded as processes close to stationary. On the basis of this hypothesis, a method of approximating isolated epidemic process waves by means of generalized logistic functions with an increased number of exponents was proposed. The developed approach was applied to predict the number of infected people in the Republic of Belarus for the period until August 2020 based on data from the beginning of the epidemic until June 12, 2020.

Multivariate Fractional Components Analysis

We propose a setup for fractionally cointegrated time series which is formulated in terms of latent integrated and short-memory components. It accommodates nonstationary processes with different fractional orders and cointegration of different strengths and is applicable in high-dimensional settings. In an application to realized covariance matrices, we find that orthogonal short- and long-memory components provide a reasonable fit and competitive out-of-sample performance compared with several competing methods.

CHOOSING A QUALITY CRITERION FOR EVALUATING THE FORECAST OF NONLINEAR NON-STATIONARY PROCESSES

Background. The problem of forecasting nonlinear nonstationary processes presented in the form of time series is very relevant, since such series can describe dynamics of the processes in both technical and economic systems. To establish the best model, various metrics are used to assess the quality of forecasts, such as R^2, RMSE, MAE, MAPE. However, in many tasks, when optimizing the model according to the selected criterion, the model becomes worse in relation to another criterion. Therefore it is important to understand which metric must be used to optimize and assess the quality of the forecast in the given task. Objective. The aim of the paper is to develop a criteria base for assessing forecasts of nonlinear nonstationary processes, as well as an approach to choosing a metric in accordance to the specificity of the set forecasting problem. Methods. The paper presents a comparative analysis of the basic metrics for the regression problem, their theoretical and practical meaning, advantages and disadvantages in various cases. New approaches are proposed based on the results of the analysis. Results. Based on the analysis of the selected data, it is shown that by optimizing the model according to the selected criterion, the model becomes worse in relation to another criterion. A criterion basis for assessing forecasts of nonlinear nonstationary processes has been formed, as well as an approach to the selection of a quality criterion in accordance with the specifics of the set forecasting problem. To minimize an absolute error, the RMSE (MSE, R^2) and MAE metrics are analysed and recommended, depending on the need to work with outliers. The RMSLE metric is proposed for solving the problems of minimizing the relative metric, for solving the shown problems of the MAPE metric for this class of problems. Conclusions. The paper shows the importance of choosing a metric that must be used to optimize and assess the quality of the forecasts in the given task. The obtained criterion base and approach can be used in further research to solve practical prob- lems in modelling and forecasting nonlinear nonstationary processes and to develop new methods or general method for solving such problems.

General methods of forecasting nonlinear nonstationary processes based on mathematical models using statistical data

The article considers the problem of forecasting nonlinear nonstationary processes, presented in the form of time series, which can describe the dynamics of processes in both technical and economic systems. The general technique of analysis of such data and construction of corresponding mathematical models based on autoregressive models and recurrent neural networks is described in detail. The technique is applied on practical examples while performing the comparative analysis of models of forecasting of quantity of channels of service of cellular subscribers for a given station and revealing advantages and disadvantages of each method. The need to improve the existing methodology and develop a new approach is formulated.

Convolutional neural networks for modeling and forecasting nonlinear nonstationary processes

The object of research. The object of research is modeling and forecasting nonlinear nonstationary processes presented in the form of time-series data. Investigated problem. There are several popular approaches to solving the problems of adequate model constructing and forecasting nonlinear nonstationary processes, such as autoregressive models and recurrent neural networks. However, each of them has its advantages and drawbacks. Autoregressive models cannot deal with the nonlinear or combined influence of previous states or external factors. Recurrent neural networks are computationally expensive and cannot work with sequences of high length or frequency. The main scientific result. The model for forecasting nonlinear nonstationary processes presented in the form of the time series data was built using convolutional neural networks. The current study shows results in which convolutional networks are superior to recurrent ones in terms of both accuracy and complexity. It was possible to build a more accurate model with a much fewer number of parameters. It indicates that one-dimensional convolutional neural networks can be a quite reasonable choice for solving time series forecasting problems. The area of practical use of the research results. Forecasting dynamics of processes in economy, finances, ecology, healthcare, technical systems and other areas exhibiting the types of nonlinear nonstationary processes. Innovative technological product. Methodology of using convolutional neural networks for modeling and forecasting nonlinear nonstationary processes presented in the form of time-series data. Scope of the innovative technological product. Nonlinear nonstationary processes presented in the form of time-series data.

DATA FILTERING TECHNIQUES IN DECISION SUPPORT SYSTEMS

Background. The majority of modern dynamic processes in economy, finances, ecology, technologies and many other areas of studies exhibit short- and long-term nonlinear and nonstationary behavior. That is why it is required to create for their thorough analysis modern highly developed specialized instrumentation providing for appropriate preliminary statistical data processing, simulation state and parameter estimation and quality forecasting their evolution in time to be used in decision support systems (DSS). Objective. The purpose of the paper is to perform introductory analysis of some modern methods for filtering statistical and experimental data; to consider modern filtering techniques on the basis of probabilistic Bayesian approach, that provide a possibility for preparing the data to adequate simulation, computing high quality state and forecast estimates for dynamic systems in stochastic environment and availability of measurement errors. Methods. To implement modern data filtering techniques appropriate simulation and optimization procedures, probabilistic Bayesian methods of data analysis are utilized; simulation algorithms for parameter estimation, and criteria bases for analyzing quality of intermediate and final results in the frames of DSS are used. Results. A set of data filtering techniques is presented to be used together with the models describing formally selected processes dynamics. The methodology is considered for implementation of probabilistic Bayesian filter based upon modern statistical data analysis techniques including application of appropriate simulation procedures. Conclusions. Development of effective means for simulation, state estimation and forecasting dynamics of nonlinear nonstationary processes in various areas of human activities provides a possibility for high quality state and parameter estimation and compute short and middle term forecasts for their future evolution. The methods of optimal Kalman and probabilistic Bayesian filtering considered in the review provide a possibility for performing appropriate analysis of nonlinear nonstationary processes, compute forecasts and provide for managerial decision support on the basis of the forecast estimates.