Bitcoin’s sentiment spillover effect and hybrid CNN-LSTM framework for cryptocurrency price forecasting
PeerJ Computer Science · 13.07.2026 · 12 · e3808
DOI: 10.7717/peerj-cs.3808 ↗Kurz gefasst
Fachartikel in der Zeitschrift PeerJ Computer Science, online erschienen am 13.07.2026. Verfasst von Amine Sehairi und 1 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Northumbria University Library.
Abstract (Originalsprache)
This article investigates how sentiments surrounding Bitcoin affect other major cryptocurrencies and uses Bitcoin sentiments along with other features to predict the price of these assets. Since Bitcoin is the market leader and several studies have shown the interconnectedness of the cryptocurrency market, understanding and quantifying the effect of Bitcoin-related sentiment on other coins can provide new insights for investors and traders, and expand the horizon of trading signals. After applying statistical tests, we found that sentiment affects cryptocurrencies differently across market regimes. In particular, Ethereum demonstrated a bidirectional relationship with Bitcoin sentiment, while Solana reacted almost instantly, and X Ripple Protocol (XRP) displayed a delayed response. Building on these results, we developed a deep learning model with a parallel convolutional neural network-long short memory (CNN-LSTM) architecture, where convolutional layers extract high-level temporal representations from multivariate market and sentiment features, and a recurrent LSTM branch models sequential dependencies. The learned representations from both components are subsequently fused to generate the final price prediction. The convolutional component was designed to capture short-term market fluctuations and abrupt sentiment-driven reactions, while the LSTM component models longer-term temporal dependencies and sustained market trends using Bitcoin sentiment, Bitcoin features, Solana features, and their interactions to predict Solana’s price. Lagged features were used in some combinations with other attributes. Through experimentation with different lookback windows and forecast horizons, Model 5 achieved the best results, recording the lowest overall errors (mean square error (MSE) = 255.35, mean absolute error (MAE) = 12.01). Notably, it employed the shortest lookback configuration (24 h CNN, 72 h LSTM). This framework, which focuses on extracting market conditions from Bitcoin sentiment and features while incorporating asset-specific reactions, offers a unique way of integrating broader market signals into predictive models of major cryptocurrencies.
Förderung
- Northumbria University Library
Zitieren
APA
Sehairi, A., & Herzog, N. J. (2026). Bitcoin’s sentiment spillover effect and hybrid CNN-LSTM framework for cryptocurrency price forecasting. PeerJ Computer Science, 12, e3808. https://doi.org/10.7717/peerj-cs.3808
Vancouver
Sehairi A, Herzog NJ. Bitcoin’s sentiment spillover effect and hybrid CNN-LSTM framework for cryptocurrency price forecasting. PeerJ Computer Science. 2026;12:e3808. doi:10.7717/peerj-cs.3808
BibTeX
@article{Sehairi2026,
author = {Sehairi, Amine and Herzog, Nitsa J.},
title = {Bitcoin’s sentiment spillover effect and hybrid CNN-LSTM framework for cryptocurrency price forecasting},
journal = {PeerJ Computer Science},
year = {2026},
volume = {12},
pages = {e3808},
doi = {10.7717/peerj-cs.3808}
}
EUVIMED