Multi-level oil price prediction based on BIGRU and LightGBM

This study proposes a hybrid model that integrates node gated recurrent unit (BiGRU) and lightweight gradient boosting machine (LightGBM) to predict international crude oil price fluctuations. The model captures the long-term correlation and dynamic characteristics of multi-dimensional control data through BiGRU, and focuses on key time nodes in combination with the focus mechanism; LightGBM integrates statistical features and node instantaneous jumps, and uses node boosting decision tree nonlinear modeling. Based on data such as WTI crude oil, US dollar index and Baltic Dry Index from 2019 to 2024, the results are better than existing methods in terms of MAE, MSE and R² evaluation indicators, showing high prediction accuracy. The predicted curve is highly consistent with the actual price trend, and the ability to capture key inflection points significantly exceeds the traditional relationship model. This study provides a topological prediction framework for the energy financial market, and will explore its extended application in the derivatives market in the future.