Research and Analysis of Cryptocurrency Price Prediction Based on Deep Learning

Cryptocurrency markets trade 24/7 and show extreme volatility, heavy-tailed returns, and regime shifts, making forecasting difficult yet valuable for risk management and trading. This review synthesizes recent AI research on cryptocurrency prediction with a focus on deep learning. The paper classifies studies into CNN-based models, RNN and hybrid architectures, attention/Transformer methods, graph neural networks capturing cross-asset effects, and decomposition or multimodal pipelines that integrate signals such as sentiment. The paper compares work by forecasting targets (price, return, direction), data modalities (OHLCV, indicators, macro stress, text), and evaluation protocols (time-aware splits and walk-forward validation). Reported improvements are often regime-dependent and highly sensitive to leakage-prone design choices, complicating fair comparison and deployment. The paper concludes with recommendations for standardized benchmarks, uncertainty-aware modeling, and strictly time-aligned multimodal inputs.