From Bitcoin to Altcoins: A Critical Review of Deep Learning Models in Cryptocurrency Price Prediction

Abstract

Cryptocurrencies such as Bitcoin and Ethereum have transformed global finance but remain difficult to forecast due to volatility, nonlinear dynamics, and sensitivity to speculative and macroeconomic shocks. Traditional econometric models provide baselines but fail to capture chaotic and high-frequency patterns. Deep learning architectures including LSTM, GRU, and Bi-LSTM offer stronger predictive capacity, yet issues of interpretability, scalability, and data heterogeneity persist.
This study presents a scientometric and thematic review of deep learning in cryptocurrency forecasting using Scopus AI Analytics, covering publications from 2010 to 2025. The analysis maps research trends, conceptual clusters, leading experts, and emerging themes. Findings indicate a methodological progression from single model Bitcoin focused studies toward hybrid and ensemble frameworks that integrate technical indicators, sentiment analysis, and blockchain-based features. Evaluation metrics have also shifted beyond error minimization toward profitability and risk-adjusted performance.
The review highlights three critical frontiers: (i) hybrid and ensemble architectures for robustness, (ii) explainable AI for interpretability, and (iii) integration of multi-source and behavioral data for practical forecasting. Collectively, these insights underscore the need for resilient, interpretable, and multi-asset deep learning systems, offering implications for both academic research and financial decision-making in volatile digital asset markets.