Scholarlink

This in an era of fluctuating oil prices that significantly impact global economies, accurate forecasting has become crucial for stakeholders in the energy sector. This study systematically compares the efficacy of three prominent deep learning models—Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Transformer-based BERT—on the task of oil price prediction. Utilizing a comprehensive dataset of historical crude oil prices, we preprocess the data to enhance model training and optimize hyper parameters to achieve robust performance. Our results indicate that the CNN model outperforms its counterparts, achieving the lowest Root Mean Squared Error (RMSE), thereby demonstrating its superior ability to capture temporal patterns in oil price dynamics. The LSTM model follows closely, showcasing its strength in handling sequential data, while the Transformer model, despite its architectural sophistication, yields slightly less accuracy in this context. The findings underscore the potential of CNN architectures in financial forecasting applications and provide valuable insights for practitioners aiming to harness machine learning techniques for effective oil price prediction.

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

Comparing RNN, LSTM, and GRU Models in Oil Price Forecasting

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hicham Boussatta,
  • Marouane Chihab,
  • Mohamed Chiny,
  • Younes Chihab

摘要

This in an era of fluctuating oil prices that significantly impact global economies, accurate forecasting has become crucial for stakeholders in the energy sector. This study systematically compares the efficacy of three prominent deep learning models—Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Transformer-based BERT—on the task of oil price prediction. Utilizing a comprehensive dataset of historical crude oil prices, we preprocess the data to enhance model training and optimize hyper parameters to achieve robust performance. Our results indicate that the CNN model outperforms its counterparts, achieving the lowest Root Mean Squared Error (RMSE), thereby demonstrating its superior ability to capture temporal patterns in oil price dynamics. The LSTM model follows closely, showcasing its strength in handling sequential data, while the Transformer model, despite its architectural sophistication, yields slightly less accuracy in this context. The findings underscore the potential of CNN architectures in financial forecasting applications and provide valuable insights for practitioners aiming to harness machine learning techniques for effective oil price prediction.