Оригинал (Original)
Автори: Копаранов, К. А., Георгиев, К. К., Щерев, В. А.
Заглавие: ВЛИЯНИЕ НА ДЪЛЖИНАТА НА ИСТОРИЯТА И БРОЯ ЕПОХИ ПРИ ПРОГНОЗИРАНЕТО НА ВРЕМЕВИ РЕД С ПОМОЩТА НА НЕВРОННА МРЕЖА С ДЪЛГА КРАТКОСРОЧНА ПАМЕТ
Ключови думи: Deep Learning; LSTM; RNN; Stock Price Predicting; Time Series

Абстракт: Forecasting the future values of time series is a problem that occurs in various subject areas such as: finance, meteorology, logistics, etc. The techniques and tools of machine learning have wide application for solving such tasks in recent years. However, they have not been sufficiently studied. The article explores the influence of the lookback period (time lag) and number of training epochs in forecasting time series using a deep neural network with long shortterm memory. It is monitored how different combinations of the above parameters affect the accuracy of the forecast, with the aim to reach an optimal combination of their values. As a result of a numerical experiment with an example financial time series, it was found that the use of more lags (over two to four) does not improve the results due to problems in model training. Such a study of the parameters of the model is important for their proper selection in solving practical problems.

Библиография

    Издание

    International Scientific Conference UNITECH’20, том I, брой 20, стр. стр. 319-324, 2020, България, Габрово, Университетско издателство “Васил Априлов” – Габрово, ISBN ISSN 1313-230X
    Autors: Koparanov, K. A., Georgiev, K. K., Shterev, V. A.
    Title: INFLUENCE OF THE LOOKBACK PERIOD AND THE NUMBER OF EPOCHS IN THE PREDICTION OF TIME SERIES USING A NEURAL NETWORK WITH A LONG SHORT-TERM MEMORY
    Keywords: Deep Learning; LSTM; RNN; Stock Price Predicting; Time Series

    Abstract: Forecasting the future values of time series is a problem that occurs in various subject areas such as: finance, meteorology, logistics, etc. The techniques and tools of machine learning have wide application for solving such tasks in recent years. However, they have not been sufficiently studied. The article explores the influence of the lookback period (time lag) and number of training epochs in forecasting time series using a deep neural network with long shortterm memory. It is monitored how different combinations of the above parameters affect the accuracy of the forecast, with the aim to reach an optimal combination of their values. As a result of a numerical experiment with an example financial time series, it was found that the use of more lags (over two to four) does not improve the results due to problems in model training. Such a study of the parameters of the model is important for their proper selection in solving practical problems.

    References

      Issue

      International Scientific Conference UNITECH’20, vol. 1, issue 20, pp. 319-324, 2020, Bulgaria, Gabrovo, UNIVERSITY PUBLISHING HOUSE “V. APRILOV” – GABROVO, 2020, ISBN ISSN 1313-230X

      Вид: публикация в национален форум с межд. уч., публикация в реферирано издание