@article{article_822153, title={Comparison of Statistical and Machine Learning Algorithms for Forecasting Daily Bitcoin Returns}, journal={Avrupa Bilim ve Teknoloji Dergisi}, pages={444–454}, year={2021}, DOI={10.31590/ejosat.822153}, author={Aygun, Betul and Kabakcı Gunay, Eylul}, keywords={Bitcoin, Forecasting, Statistical Analysis, Machine Learning, DNN, RNN, LSTM, MVA, ARIMA}, abstract={Increasing fluctuations in pricing and having great potential for profit in finance world, utilization in advanced machine learning technologies to make robust prediction of highly nonlinear prices of cryptocurrencies especially Bitcoin have attracted great attention in recent years and forecasting its value have become an interesting research subject in different areas such as computer science and finance. In this study, various statistical and machine learning techniques have been conducted and compared to predict Bitcoin crypto currency price. By using the historical daily Bitcoin data between January 1, 2013 to October 15, 2020, Moving Average Analysis and Autoregressive Integrated Moving Average among the statistical techniques, and among Machine Learning (ML) techniques Neural network, Deep neural network, Recurrent Neural Network (RNN) and LSTM, one of the most preferred RNN have been applied for the univariate time series analysis with window size of 32. It has aimed to justify the usefulness of the machine learning algorithms instead of just apply statistical techniques. Besides, Bitcoin value of the first month of the 2021 has been predicted. To prove the usefulness of ML algorithms, and to show that deeper algorithm results better, mean squared error (MSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) metrics have been used for comparison. The results have revealed that deep learning algorithm yields better than other methods in predicting daily Bitcoin price in terms of MSE, MAE and MAPE metrics proportional to the deepness of the algorithms. Nonetheless, LSTM does not have capability to train the network according to the parameters selected during this study and it has not been outperformed than ARIMA statistical method and other ML algorithms.}, number={21}, publisher={Osman SAĞDIÇ}