Resumen

LAMBIS-ALANDETE, Erick; JIMENEZ-GOMEZ, Miguel  y  VELASQUEZ-HENAO, Juan D.. Comparison of Deep Learning algorithms for cryptocurrencies price forecasting. Ing. compet. [online]. 2023, vol.25, n.3, e-21312845.  Epub 26-Jun-2023. ISSN 0123-3033.  https://doi.org/10.25100/iyc.v25i3.12845.

Due to the growth and interest that cryptocurrencies have generated nowadays, investors and researchers have paid more attention to forecasting the prices of cryptocurrencies. With the methodological development of Deep Learning, forecasting cryptocurrencies has become more important in recent years. In this paper, four Deep Learning models RNN, LSTM, GRU and CNN-LSTM are evaluated with the aim of evaluating the performance in forecasting the daily closing price of the two most important cryptocurrencies: Bitcoin and Ethereum. Performance analysis metrics such as MAE, RMSE, MSE and MAPE were used and as a fitting metric, the R2. Each Deep Learning model was optimized from a set of hyperparameters and for different time windows. Experimental results showed that the RNN algorithm had superior performance in predicting the Bitcoin price and the LSTM algorithm in predicting the Ethereum price. Even, both methods presented better performance with two literature models evaluated. Finally, the forecast reliability of each model was evaluated by analyzing the autocorrelation of the errors and the two most efficient models were found to have high generalization power.

Palabras clave : deep Learning; Cryptocurrencies; Time Series; investor.

        · resumen en Español     · texto en Español     · Español ( pdf )