Designing a Predictive and Explainable Model for Stablecoin Prices by Integrating On-Chain Data and Deep Learning-Based Market Sentiment Analysis

Mohammadsaeid  Daliri; Mohammadreza  Sanaei; Roozbeh  Balounejad Nouri

Authors

Mohammadsaeid Daliri Department of IT Management, QA.C., Islamic Azad university , Qazvin, Iran Author https://orcid.org/0009-0002-2460-096X
Mohammadreza Sanaei Department of IT Management, QA.C., Islamic Azad university , Qazvin, Iran https://orcid.org/0009-0007-1653-0718
Roozbeh Balounejad Nouri Department of Economic Science, QA.C., Islamic Azad university, Qazvin, Iran Author https://orcid.org/0000-0002-9355-2554

Keywords:

Stablecoin, Price Prediction, Deep Learning, Attention, LSTM, On, Chain Data, Market Sentiment, Explainable Artificial Intelligence, Cryptocurrency

Abstract

This study aimed to design and evaluate an explainable deep learning model for predicting stablecoin prices and peg deviations by integrating on-chain indicators, conventional market variables, and market sentiment data. This applied predictive-analytical study was conducted in Tehran, Iran, during 2025–2026. The behavioral validation sample consisted of 384 active cryptocurrency market participants. The longitudinal dataset included 2,190 synchronized daily observations for USDT, USDC, and DAI. Data comprised historical prices, trading characteristics, transaction volume, active addresses, exchange inflows and outflows, net exchange flows, large-holder transactions, supply changes, and deep learning-derived sentiment indicators. Sentiment was extracted from cryptocurrency-related textual data using transformer-based natural language processing. An integrated Attention-LSTM model was developed and compared with Support Vector Regression, Random Forest, Extreme Gradient Boosting, conventional LSTM, and GRU models. Chronological training, validation, and test partitions were used, and SHAP analysis assessed feature importance and direction. The integrated Attention-LSTM model demonstrated the strongest out-of-sample performance, with MAE = 0.00196, RMSE = 0.00324, MAPE = 0.195%, R² = 0.939, and directional accuracy = 83.23%. It outperformed the Attention-LSTM model without sentiment, which achieved RMSE = 0.00453 and R² = 0.881. Ablation analysis showed that the full integrated model reduced RMSE by 53.31% relative to the historical-price-only model. The model achieved R² values of 0.947, 0.942, and 0.921 for USDT, USDC, and DAI, respectively, and directional accuracy reached 90.32% during high-deviation periods. SHAP analysis identified negative sentiment intensity, net exchange inflow, previous-day peg deviation, on-chain transaction volume, sentiment dispersion, and large-holder transaction volume as the most influential predictors. Integrating on-chain, market, and sentiment information within an explainable Attention-LSTM framework substantially improves stablecoin price forecasting and provides interpretable early-warning signals of peg instability.

Downloads

Download data is not yet available.

References

1. Zhang Y, Lee J. Deep Learning for Cryptocurrency Price Prediction: A Survey. Journal of Financial Data Science. 2020;2(3):15-35.

2. Zhou J, Li H. Time Series Analysis of Cryptocurrency Prices Using Deep Learning. Journal of Computational Finance. 2021;25(1):56-72.

3. Huang Y, Zhao M. Cryptocurrency Price Prediction Using Hybrid Deep Learning Techniques. Expert Systems with Applications. 2019;123:195-210.

4. Patel R, Verma A. Cryptocurrency Price Prediction Using Ensemble Deep Learning Models. Expert Systems with Applications. 2021;183:115392.

5. Pintelas E, Livieris IE, Stavroyiannis S, Kotsilieris T, Pintelas P. Investigating the Problem of Cryptocurrency Price Prediction: A Deep Learning Approach. AIAI 2020: Artificial Intelligence Applications and Innovations2020. p. 99-110.

6. Bullmann D, Klemm J, Pinna A. In search for stability in crypto-assets: Are stablecoins the solution? European Central Bank Occasional Paper Series, 2019.

7. Barucci E, Moncayo GG, Marazzina D. Cryptocurrencies and Stablecoins: A High-Frequency Analysis. Digital Finance. 2022;4(2-3):217-39. doi: 10.1007/s42521-022-00055-9.

8. Bouteska A, Abedin MZ, Hajek P, Yuan K. Cryptocurrency price forecasting - A comparative analysis of ensemble learning and deep learning methods. International Review of Financial Analysis. 2024;92:103055. doi: 10.1016/j.irfa.2023.103055.

9. Akouaouch I, Bouayad A. A new deep learning approach for predicting high -frequency short -term cryptocurrency price. Bulletin of Electrical Engineering and Informatics. 2025;14(1):513-23. doi: https://doi.org/10.11591/eei.v14i1.7377.

10. Bakhit S, Rabiei M. Forecasting the Price of the First Virtual Cryptocurrency Based on Deep Machine Learning Techniques: Ghiyasuddin Jamshid Kashani University; 2020.

11. Roberts J. Sentiment Analysis and Cryptocurrency Price Prediction: A Deep Learning Approach. Journal of Financial Markets. 2018;37:85-101.

12. Rusydiana AS. Cryptocurrency: A Sentiment Analysis From Twitter Data. Ift. 2024;1(1). doi: 10.58968/ift.v1i1.359.

13. Farzulla M. The Extremity Premium: Sentiment Regimes and Adverse Selection in Cryptocurrency Markets. 2026. doi: 10.48550/arxiv.2602.07018.

14. Sakariyahu R, Lawal R, Adigun R, Paterson A, Johan S. One Crash, Too Many: Global Uncertainty, Sentiment Factors and Cryptocurrency Market. Journal of International Financial Markets, Institutions and Money. 2024;94. doi: 10.1016/j.intfin.2024.102028.

15. Nourollahi S, Jafari A, Pakmaram A. The Impact of Investor Sentiment on Portfolio Optimization in the Tehran Stock Exchange and Cryptocurrency Markets. Business, Marketing, and Finance Open. 2025:1-21.

16. Soltani H, Taleb J, Boujelbene Abbes M. Spillover effects between stock markets, cryptocurrency, and investor sentiment during COVID-19. European Journal of Management and Business Economics. 2025;34(1):23-46.