Maryam Seifaddini - Assistant Professor of Computer Science at University of Guilan, Rasht, Iran;
Amir Habibdoust - Lecturer at University of Guilan, Rasht, Iran;

In this paper, we aim to compare the Bitcoin price forecasting performance of a standard machine learning method by looking at data from two different lenses. Indeed, we combine the standard GMDH-neural network with wavelet analysis. In this regard, we decomposed Bitcoin price over the period 2018/12/27 to 2020/08/26 on daily basis at 6 scales and predict the price based on input at different time-scales spaces. Finally, the data are summed up and transformed into time-domain space. Due to differences between features and price movements of Cryptocurrencies and other financial assets, it would be valuable if traders know about the importance of trade frequencies' impact on forecasting power. The results show a better performance of the wavelet base GMDH-neural network in comparison with the standard method. Moreover, it means that there is valuable information on different scales of Bitcoin prices which are neglected in standard time series.

Machine Learning, Wavelet, Bitcoin Price, Forecasting,