Scholarlink

<p>Accurate and reliable forecasting of Bitcoin prices is important for market participants to obtain potentially high returns and make efficient decisions. To improve the forecasting performance of Bitcoin prices, this paper uses a novel deep hybrid model, viz., EMD-CNN-GRU, that combines empirical mode decomposition (EMD), convolutional neural network (CNN), and gated recurrent unit (GRU), to generate a one-step-ahead forecast of Bitcoin prices. The deep hybrid model exploits the advantage of empirical mode decomposition for decomposing noise series into their respective intrinsic mode functions and one residue, the ability of convolutional neural network for extracting local valuable features and learning the internal representation of each sub-sequence obtained from empirical mode decomposition, as well as the effectiveness of gated recurrent unit for identifying long-term dependencies in time series data and automatically detecting the best mode suitable for relevant data. Our findings demonstrate that the EMD-CNN-GRU model obtains higher forecasting accuracy of Bitcoin prices than other counterparts according to the commonly used metrics for time series forecasting. The empirical results exhibit compelling evidence that the suggested deep hybrid model based on EMD and multiple deep learning networks sheds light on the essential features and price changing trends of Bitcoin and provides reliable and practical experience for market participants to study the predictability of cryptocurrencies inclusive of Bitcoin.</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

Bitcoin price forecasting with a deep hybrid EMD-CNN-GRU model

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Xunfa Lu,
  • Jinghao Shao,
  • Kin Keung Lai,
  • Yuhong Shi,
  • Qian Chen,
  • Hairong Cui

摘要

Accurate and reliable forecasting of Bitcoin prices is important for market participants to obtain potentially high returns and make efficient decisions. To improve the forecasting performance of Bitcoin prices, this paper uses a novel deep hybrid model, viz., EMD-CNN-GRU, that combines empirical mode decomposition (EMD), convolutional neural network (CNN), and gated recurrent unit (GRU), to generate a one-step-ahead forecast of Bitcoin prices. The deep hybrid model exploits the advantage of empirical mode decomposition for decomposing noise series into their respective intrinsic mode functions and one residue, the ability of convolutional neural network for extracting local valuable features and learning the internal representation of each sub-sequence obtained from empirical mode decomposition, as well as the effectiveness of gated recurrent unit for identifying long-term dependencies in time series data and automatically detecting the best mode suitable for relevant data. Our findings demonstrate that the EMD-CNN-GRU model obtains higher forecasting accuracy of Bitcoin prices than other counterparts according to the commonly used metrics for time series forecasting. The empirical results exhibit compelling evidence that the suggested deep hybrid model based on EMD and multiple deep learning networks sheds light on the essential features and price changing trends of Bitcoin and provides reliable and practical experience for market participants to study the predictability of cryptocurrencies inclusive of Bitcoin.