| Keywords [eng] |
time series, bitcoin, ARIMA, VAR, VARMA, SVM, LightGBM, XGBoost, Random Forest, Decision Tree, Adaboost, blending, time-series forecasting, machine learning, cryptocurrency, permutation feature importance, Extra Trees, cross-validation, feature selection |
| Abstract [eng] |
In recent years, with the increase in the prevalence of cryptocurrencies, the interest in the problem of predicting cryptocurrencies has also started to increase. This study aims to predict Bitcoin's return using different machine learning and time series models. Different training data intervals, cross-validation, feature engineering, and various statistical tests are used to achieve high performance in each model. The training data is taken from Binance API. The results show that machine learning algorithms outperform time-series models. Machine learning algorithms blending helps to obtain even better results. |
