Predicting Cryptocurrency Price Dynamics: A Comparative Analysis of Machine Learning Techniques
Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi · 31.05.2026 · 13 (1) · 17-35
DOI: 10.35193/bseufbd.1724604 ↗Kurz gefasst
Fachartikel in der Zeitschrift Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi, online erschienen am 31.05.2026. Verfasst von Muhammed Allito und 1 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben.
Abstract (Originalsprache)
In recent years, the rapid acceleration of digitalization has led to significant transformations in financial systems, with cryptocurrencies drawing attention due to their decentralized structures, increasing transaction volumes, and investment potential. However, high price volatility and market uncertainties make cryptocurrency price prediction challenging, thereby increasing the strategic importance of developing effective predictive models in this field. In this study, the prices of 15 popular cryptocurrencies were comparatively evaluated using four regression algorithms—Gradient Boosting Regression (GBR), Random Forest Regression (RFR), Ridge Regression (RR), and Light Gradient Boosting Machine (LGBM)—at hourly and daily time resolutions. Two distinct modeling strategies were adopted: the first involved training independent models for each cryptocurrency, while the second combined all asset data into multivariate models. In the individual datasets, the RR algorithm achieved the highest accuracy in capturing short-term fluctuations (R² ≈ 0.998, MAE ≈ 0.007, RMSE ≈ 0.008), while GBR and RFR demonstrated competitive performance. The LGBM model exhibited higher sensitivity to short-term volatility for certain assets. In the multivariate datasets, RFR produced the most stable and accurate predictions for both hourly and daily data (MAE ≈ 0.038 hourly, 0.165 daily; RMSE ≈ 0.185 hourly, 0.53 daily). While LGBM maintained strong performance on short-term data and GBR remained generally consistent—albeit with higher hourly RMSE for some assets—RR produced larger errors in certain daily scenarios. The results indicate that the success of cryptocurrency price prediction depends not only on the chosen algorithm but also strongly on the data structure, modeling strategy, and temporal resolution; short-term price movements can be predicted with high accuracy using both individual and multivariate datasets, whereas in long-term forecasts, algorithm selection and data integration play a decisive role.
Zitieren
APA
Allito, M., & Çelik, E. (2026). Predicting Cryptocurrency Price Dynamics: A Comparative Analysis of Machine Learning Techniques. Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi, 13(1), 17-35. https://doi.org/10.35193/bseufbd.1724604
Vancouver
Allito M, Çelik E. Predicting Cryptocurrency Price Dynamics: A Comparative Analysis of Machine Learning Techniques. Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi. 2026;13(1):17-35. doi:10.35193/bseufbd.1724604
BibTeX
@article{Allito2026,
author = {Allito, Muhammed and Çelik, Esra},
title = {Predicting Cryptocurrency Price Dynamics: A Comparative Analysis of Machine Learning Techniques},
journal = {Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi},
year = {2026},
volume = {13},
number = {1},
pages = {17--35},
doi = {10.35193/bseufbd.1724604}
}
EUVIMED