Combination of several algorithms while solving the forecasting problem
Journal: The Journal of Zhytomyr State Technological University. Series: Engineering (Vol.1, No. 76)Publication Date: 2016-07-15
Authors : O.I. Chumachenko; V.S. Horbatiuk;
Page : 101-107
Keywords : time series forecasting; combination of models; artificial neural networks; group method of data handling;
Abstract
The forecasting problem is undoubtedly one of the most important for humanity, but also it is standing next to the most difficult ones because there is no guarantee that the predicted process behavior will not change dramatically in the future. The aim of the paper is to develop a new forecasting method that uses an approach of several models combining and test its quality on a set of test data. For this purpose, the main existing forecasting methods were considered, such as artificial neural networks, group method of data handling and linear regression. For the combination of forecasting models the bagging approach was chosen. The proposed method was implemented in a software environment Matlab and tested (along with several existing methods) on 11 sets of test data. The fact that among the methods that were tested, the proposed one showed the best results, indicating the possibility of its successful application in practice is a scientific novelty and practical significance of the article.
Other Latest Articles
- Thesaurus of accounting science: historical path of formation and present identification of ?calculation? concept
- To the question of increasing the accuracy of angular measurements of goniometric system
- Forecasting of enterprise’s financial indicators as the basis for financial stability formation
- Оbjects of qualitative approach in accounting
- Experimental research of the frequency dependence of microwave radiation coefficients of reflection and transmission in the nanometer conductive inclusions
Last modified: 2016-09-29 18:02:50