Scholarlink

<p>Cryptocurrencies have experienced exponential growth, with over 25,000 cryptocurrencies amassing a combined market value of more than $2 trillion as of 2024. Bitcoin, the pioneer and most prominent cryptocurrency, continues to attract investors aiming to maximize profits while minimizing risks. However, accurately forecasting Bitcoin’s future price remains a significant challenge due to its extreme volatility. Existing research largely relies on limited datasets, inadequate pre-processing techniques, and models that lack learning of complex temporal patterns and long-term dependencies, leading to suboptimal predictions. In this paper, we address these critical gaps by proposing a novel and comprehensive framework for Bitcoin price prediction, leveraging all available historical data from 2009 to 2024. Our approach introduces a robust pre-processing pipeline, including advanced feature selection, outlier detection, and data normalization, which has been widely overlooked in previous studies. By improving data quality and reducing noise, we ensure that our model captures real-world market dynamics. Additionally, we implement a deep learning-based stacked-GRU model, capable of learning complex temporal patterns and long-term dependencies. Our model’s superior performance is validated through rigorous evaluation against state-of-the-art models, achieving an <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> of 0.998, MSE of 0.0003, RMSE of 0.0185, and MAE of 0.0102. These results highlight the significant flaws in existing research, which often neglects critical data after 2020 and fails to account for the inherent volatility of Bitcoin price movements.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced data quality-driven framework using stacked-GRU for robust prediction of volatile crypto trends

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Muhammad Wajid,
  • Ahmad Kamran Malik,
  • Nafees Qamar,
  • Muneeba Islam,
  • Iftikhar Azim Niaz

摘要

Cryptocurrencies have experienced exponential growth, with over 25,000 cryptocurrencies amassing a combined market value of more than $2 trillion as of 2024. Bitcoin, the pioneer and most prominent cryptocurrency, continues to attract investors aiming to maximize profits while minimizing risks. However, accurately forecasting Bitcoin’s future price remains a significant challenge due to its extreme volatility. Existing research largely relies on limited datasets, inadequate pre-processing techniques, and models that lack learning of complex temporal patterns and long-term dependencies, leading to suboptimal predictions. In this paper, we address these critical gaps by proposing a novel and comprehensive framework for Bitcoin price prediction, leveraging all available historical data from 2009 to 2024. Our approach introduces a robust pre-processing pipeline, including advanced feature selection, outlier detection, and data normalization, which has been widely overlooked in previous studies. By improving data quality and reducing noise, we ensure that our model captures real-world market dynamics. Additionally, we implement a deep learning-based stacked-GRU model, capable of learning complex temporal patterns and long-term dependencies. Our model’s superior performance is validated through rigorous evaluation against state-of-the-art models, achieving an \(R^2\) of 0.998, MSE of 0.0003, RMSE of 0.0185, and MAE of 0.0102. These results highlight the significant flaws in existing research, which often neglects critical data after 2020 and fails to account for the inherent volatility of Bitcoin price movements.