Scholarlink

Accurate short-term traffic speed prediction plays a vital role in smart transportation systems especially in fast-changing city environments. This study introduces a hybrid model that blends BiLSTM and Gradient Boosting through a LightGBM fusion layer where the BiLSTM tracks time-based patterns in both directions and the boosting part handles complex residuals often missed by deep learning alone. LightGBM then combines their outputs in a flexible way that adjusts as traffic conditions change. The model was tested on the METR-LA dataset using one-hour windows and five-minute steps and it consistently outperformed both traditional and newer deep learning models showing lower average errors and strong reliability. These results show that combining temporal learning with residual correction makes the system more stable and accurate which supports its use in real-time smart traffic applications.

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

Traffic Forecasting Optimization Using Ensemble Learning: A Fusion-Based BiLSTM and Gradient Boosting Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mohammed El Kaim Billah,
  • Mohammed Alamine El Houssaini,
  • Abdelfettah Mabrouk,
  • Abdelali Hadir,
  • Souad El Houssaini

摘要

Accurate short-term traffic speed prediction plays a vital role in smart transportation systems especially in fast-changing city environments. This study introduces a hybrid model that blends BiLSTM and Gradient Boosting through a LightGBM fusion layer where the BiLSTM tracks time-based patterns in both directions and the boosting part handles complex residuals often missed by deep learning alone. LightGBM then combines their outputs in a flexible way that adjusts as traffic conditions change. The model was tested on the METR-LA dataset using one-hour windows and five-minute steps and it consistently outperformed both traditional and newer deep learning models showing lower average errors and strong reliability. These results show that combining temporal learning with residual correction makes the system more stable and accurate which supports its use in real-time smart traffic applications.