Akademik Veri Yönetim Sistemi
YTU Graduate School of Science and Engineering GradColloquium 2024 - Artificial Intelligence, İstanbul, Türkiye, 04 Haziran 2024, ss.50-51
In this study, an analysis is conducted using five different machine learning methods for price predictions of various cryptocurrencies. Considering the increasing importance of cryptocurrencies in financial markets, the value of such an analysis is emphasized. It has been stated that machine learning methods could be effective tools for predicting price movements in cryptocurrency markets. In this study, two algorithms dependent on time series, namely ARIMA and LSTM, along with three gradient boosting algorithms, namely XGBoost, LightGBM, and CatBoost, have been employed. Three different time series (15 minutes, 1 hour, and 4 hours) have been examined with indicators for ten different cryptocurrencies (XRPUSDT, TRXUSDT, NEOUSDT, LTCUSDT, ETHUSDT, DOGEUSDT, DASHUSDT, BTCUSDT, BNBUSDT, and ADAUSDT). For each cryptocurrency, the prediction results of five different algorithms have been compared, and their performances have been evaluated using the Mean Squared Error (MSE) measurement. While no significant variations were observed in models for different time series, cryptocurrencies with lower closing values were found to yield better results for all algorithms. Considering the overall performance aspect, it is concluded that LightGBM is the algorithm with the best performance for all cryptocurrencies.