Comparative analysis of time series forecasting models effectiveness by means of a multi-criteria procedure on the example of financial indicators
Journal Title: Вісник Житомирського державного технологічного університету. Серія: Технічні науки - Year 2019, Vol 83, Issue 1
Abstract
In the course of the research, a program complex was developed for forecasting time series using models of different classes and analysis of their effectiveness. Known time series models, such as neural networks, fuzzy logic, adaptive prediction, autoregressive-moving average models, group method of data handling, singular spectral decomposition were constructed. To fully assess the suitability of a model to forecast a particular series, it is important to evaluate the model on several parameters, namely, the accuracy of the built predictions, the complexity of mathematical calculations and the adequacy of the results. Setting aside a certain criterion may lead to loss of quality of forecasting regarding the future behavior of the process. Insufficiently accurate models will give too high value of forecast error. Ignoring the complexity of a model can lead to the use of excessive number of parameters or the choice of models with an unreasonably complex architecture. The adequacy of the model explains how well the model describes the investigated process. The procedure developed by the authors allows taking into account all of the above-mentioned factors. Accuracy is measured by calculating the values of sum square and absolute error predictions. The complexity of the model is estimated using information quality criteria. Adequacy is determined by calculating the value of the adjusted determination coefficient, conducting Fisher test and analyzing the residues. The offered algorithm allows carrying out the analysis of efficiency of the constructed forecasting models. The results of his work can be used both for the direct forecasting of the values of future levels of a series, and for further studies in the field of analysis of time series. Preliminary step when constructing ensembles of prediction models or identifying emissions in a time sequence may be an example.
Authors and Affiliations
А. О. Dolhikh, O. G. Baibuz
Keywords
Related Articles
- EP ID EP622615
- DOI -
- Views 133
- Downloads 0
Predicting system for the estimated cost of real estate objects development using neural networks
Automation of user workflows is an integral part of the development of modern information and software systems. Many specialists in various subject areas perform most of their daily tasks using computer technology. This...
Determining the optimal loads on mechanism nodes using digital technology prototype
This article gives us important details about applying of a digital prototype for the purposes of reduction drive assembly testing. Working with digital prototype helps every user to make detailed virtual research, to es...
Modern approaches to the modeling the external forms of the car in a 3D environment
Today, the initial stage of creating new or modernizing the existing external surfaces of the car body is the process of forming, modeling and testing the developed models.. Therefore, it is extremely important realistic...
Study of ultrasonic exposure effect on retaining ability of diesel fuel in ammonium nitrate
The economic situation in Ukraine depends to a large extent on the mining and metallurgical complex, and therefore the introduction of safe and effective explosives for mining of minerals through the abandonment of TNT a...
Grinding gears by CBN wheels
The analysis of the results of comparing different methods of grinding conducted in the research showed that the method of polishing with two platelets is the least productive since it is intended for grinding high-preci...