G.О. Kravtsov, А.N. Prymushko, V.І. Koshell
Èlektron. model. 2020, 42(4):03-14
https://doi.org/10.15407/emodel.42.04.003
ABSTRACT
The authors propose an approach to constructing a time series forecasting model due to the synergy of autoregressive and neural network models. The authors put forward a number of requirements and conditions for the developed model. Among the requirements put forward, the central place is occupied by the requirement to build a model without the participation of a machine learning specialist. Among the conditions put forward to the time series, it should be noted the differentiability of the first order, which makes it possible to reduce the non-stationary series to the stationary one. The article describes the learning mechanism with detailed mathematical explanations. The approach outlined in the article is conceptual.
KEYWORDS
time series, forecasting, model, autoregression, neural networks, learning mechanism.
REFERENCES
- Ryabenko, E. (2016), “Automatic time series forecasting”, Tsikllektsiy «Data Mining in Action», available at: https://www.youtube.com/watch?v=u433nrxdf5k (accessed June 9, 2020).
- Sean, T. (2015), “Automatic Forecasting in Scale”, Join Statistical Meetup, available at: https://www.slideshare.net/seanjtaylor/automatic-forecasting-at-scale (accessed June 9, 2020).
- Chuchuyeva, I.A. (2012), “Time Series Prediction Model for Maximum Credibility Sampling”, Abstract of Cand. Sci. (Tech.) dissertation, 2012, Moscow State Technical University N.E.Bauman, Moscow, Russia.
- Bakanova, S.A. and Silkina, G.Yu. (2015), “Knowledge Evolution: Modeling and Applied Analysis”, Nauchno-tekhnicheskiyevedomosti Sankt-Peterburgskogogosudarstvennogopolitekhnicheskogouniversiteta. Ekonomicheskiyenauki, pp. 173-182, St. Petersburg, Russia, available at: https://cyberleninka.ru/article/n/evolyutsiya-znaniy-modelirovanie-i-prikladnoy- analiz/viewer (accessed June 9, 2020).
https://doi.org/10.5862/JE.233.18