An innovative hybrid LightGBM-BPNN model for enhanced commodity forecasting accuracy

• Hybrid LightGBM-BPNN model enhances accuracy for commodity price forecasting. • Combines linear and nonlinear modeling for robust market trend predictions. • Features like RSI and moving averages improve forecast reliability significantly. • SHAP reveals key predictors like moving averages and RSI for price trends. • Model excels under varied conditions, enhancing medium-to-long-term accuracy. Forecasting financial assets in volatile markets, like orange juice futures (OJ= F ), poses challenges due to complex price movements. This study introduces a hybrid model that combines a Light Gradient Boosting Machine (LightGBM) with a Backpropagation Neural Network (BPNN) to enhance OJ= F price predictions. The model effectively captures market dynamics by incorporating financial features and technical indicators. Random Forest (RF) and Support Vector Regression (SVR) are benchmarks. Results indicate the LightGBM-BPNN hybrid consistently outperforms others, especially in long-term forecasts, while SVR underperforms. SHAP dependence plots further validate the hybrid model's robustness, supporting the potential of hybrid models in commodity market predictions.