Scholarlink

<p>The cryptocurrency market is highly volatile, and these fluctuations, along with its complex infrastructure, make it difficult to predict using traditional methods. Furthermore, even if predictions are made, they may not reach the optimal accuracy. Moreover, the notable fluctuations in cryptocurrencies lead to changes in decision-making and confusion among investors, increasing the risk of their transactions. This study examines the potential of combining CNN and Bi-LSTM for predicting the price of Ethereum cryptocurrency using historical data. According to the results of this study, the proposed algorithm performs better than individual algorithms such as MLP, CNN, LSTM, Bi-LSTM, and GRU with mean absolute error of 0.006314, mean absolute percentage error of 1.6327, mean squared error of 0.0000963, root mean squared error of 0.009818, mean absolute scaled error of 0.993611, and R-squared (R<sup>2</sup>) of 0.937826. To ensure the effectiveness of the proposed method, several other methods proposed in other studies were also implemented and evaluated using the same data for a fair comparison. Overall, the results of this paper are better than the baseline models.</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

Proposing a hybrid CNN Bi-LSTM model for ethereum cryptocurrency price prediction

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • Ali Mansourabady,
  • Fatemeh Tabe,
  • Amir Hossein Rasekh,
  • Ali Ghermezian

摘要

The cryptocurrency market is highly volatile, and these fluctuations, along with its complex infrastructure, make it difficult to predict using traditional methods. Furthermore, even if predictions are made, they may not reach the optimal accuracy. Moreover, the notable fluctuations in cryptocurrencies lead to changes in decision-making and confusion among investors, increasing the risk of their transactions. This study examines the potential of combining CNN and Bi-LSTM for predicting the price of Ethereum cryptocurrency using historical data. According to the results of this study, the proposed algorithm performs better than individual algorithms such as MLP, CNN, LSTM, Bi-LSTM, and GRU with mean absolute error of 0.006314, mean absolute percentage error of 1.6327, mean squared error of 0.0000963, root mean squared error of 0.009818, mean absolute scaled error of 0.993611, and R-squared (R2) of 0.937826. To ensure the effectiveness of the proposed method, several other methods proposed in other studies were also implemented and evaluated using the same data for a fair comparison. Overall, the results of this paper are better than the baseline models.