Cryptocurrency Price Forecasting using LSTM: A Review

Abstract

The rapid growth and volatility of cryptocurrency markets have driven significant interest in accurate price forecasting models. Long Short-Term Memory (LSTM) networks, a type of recurrent neural network (RNN), have emerged as a powerful tool for time-series prediction due to their ability to capture long-term dependencies in sequential data. This review explores recent advancements in cryptocurrency price forecasting using LSTM models, comparing methodologies, datasets, preprocessing techniques, and performance metrics. Key challenges such as overfitting, data noise, and market unpredictability are also discussed. The paper concludes by highlighting research gaps and proposing directions for future development in LSTM-based crypto forecasting.

Introduction

Bitcoin and Ethereum have transformed finance by offering decentralized, transparent alternatives to fiat currencies. However, predicting cryptocurrency prices is challenging due to market volatility and complex nonlinear patterns, which traditional statistical methods often fail to capture. Deep learning, especially Long Short-Term Memory (LSTM) networks, has shown promise in forecasting cryptocurrency prices because of their ability to model sequential and time-dependent data.

This review paper surveys recent research on using LSTM and related models for cryptocurrency price prediction, highlighting different architectures, data processing techniques, evaluation metrics, and comparative performances. It notes that LSTM-based models generally outperform traditional methods and other machine learning approaches like ARIMA and Random Forest, particularly for volatile cryptocurrencies like Bitcoin.

Most studies use historical price data from sources like CoinMarketCap, applying preprocessing steps such as normalization and label shifting, and splitting data into training and test sets (often 80:20). Models are trained using optimizers like Adam and evaluated by metrics including RMSE and R².

Key challenges identified include the neglect of external factors (regulation, market psychology), lack of real-time high-frequency models, overfitting risks, limited hybrid and comparative studies, and insufficient multivariate and long-term analyses. Ethical considerations in algorithmic trading are also underexplored.

The paper suggests future research should integrate external and behavioral data, develop adaptive real-time forecasting systems, improve model robustness, explore hybrid modeling approaches, design multivariate and cross-market frameworks, and address long-term trends and ethical issues.

Conclusion

The review paper examined the existing state of cryptocurrency price prediction with deep learning models including LSTM, GRU, Bi-LSTM, and hybrid strategies. Although such models demonstrate encouraging performance in dealing with the non-linear and volatile nature of cryptocurrency prices, major challenges still persist, such as inadequate incorporation of external variables (such as sentiment and regulations), real-time prediction limitations, model interpretability, and inadequate attention to ethical aspects. The survey indicates that there is a call for more potent, interpretable, and hybrid models, real-time and multivariate data sources, and the analysis of behavior and cross-market effects. Fulfilling these deficits can result in more accurate and trustworthy forecasting systems that facilitate well-informed decision-making in the dynamic and fast-changing realm of digital finance.

References

[1] M. Z. A. P. H. K. Y. Ahmed Bouteska, “Cryptocurrency price forecasting – A comparative analysis of ensemble learning and deep learning methods,” International Review of Financial Analysis, vol. 92, 2024. [2] O. T. A. Ernes Korkmaz, “Cryptocurrency Price Forecasting Using LSTM with Short Time Series Data,” International Journal of Advanced Natural Sciences and Engineering Researches, vol. 8, no. 2, pp. 367-376, 2024. [3] C. R. B. M. E. P. Phumudzo Lloyd Seabe, “Forecasting Cryptocurrency Prices Using LSTM, GRU, and Bi-Directional LSTM: A Deep Learning Approach,” MDPI, 2023. [4] D. C. a. S. K. Saroj, “Cryptocurrency Price Prediction Using Supervised Machine Learning Algorithms,” ADCAIJ: Advances in Distributed Computing and Artificial Intelligence Journal, vol. 12, no. 1, 2023. [5] S. B. N. K. S. G. M. S. S. Tushar Maheshwari, “Bitcoin Price Prediction Using LSTM,” in ICCS: International Conference on Computing Sciences, 2024, 2023. [6] L. Pan, “Cryptocurrency Price Prediction Based on ARIMA, Random Forest and LSTM Algorithm,” BCP Business & Management,, vol. 38, 2023.

Copyright

Copyright © 2025 Prof. Neha Verma, Rittik Kumbhakar, Navneet Narayan, Sachin Dongre, Ajit Jadhav. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.