Application of Transfer Learning to Predict Prices of Emerging Cryptocurrencies

This survey explores the application of transfer learning techniques in cryptocurrency price prediction, particularly for emerging coins with limited historical data. Given the high volatility and non-linear patterns inherent in cryptocurrency markets, traditional forecasting models often struggle with accuracy. Transfer learning, however, offers a promising solution by enabling predictive models to leverage patterns learned from established coins, such as Bitcoin and Ethereum, for improved prediction of newer cryptocurrencies. We review recent advancements, including transformer-based models, hybrid deep learning approaches, and probabilistic methods that integrate technical indicators, social sentiment, and external market factors. Key findings highlight that models combining transfer learning with techniques like Bidirectional Long Short-Term Memory (BiLSTM), Probabilistic Gated Recurrent Units (P-GRU), and hybrid architectures significantly enhance prediction accuracy. Additionally, the survey identifies gaps in empirical research for emerging cryptocurrencies and calls for more scalable, adaptable frameworks. By synthesizing these methods and their outcomes, this survey provides insights into current trends and proposes future research directions aimed at refining transfer learning applications for robust, real-time cryptocurrency forecasting.