A differential evolution optimized hybrid XGBoost for accurate carbon emission prediction

Predicting carbon emissions is essential for combating climate change and supporting green development. Carbon emissions are influenced by complex factors, such as economy, population and new energy generation. Traditional methods struggle with these uncertainties, while machine learning offers data-driven solutions. However, some models lack data selection strategies, resulting in the neglect of critical features. To tackle this issue, this paper proposes a Differential Evolution Optimized Hybrid XGBoost (DEOH-XGBoost) approach. DEOH-XGBoost includes three main components: feature engineering, model construction, and model integration. First, in each correlation analysis, features are selected through fuzzy membership functions. Second, XGBoost-based models are constructed on each feature set to predict separately. Third, the models are integrated by a differential evolution optimized weighting strategy. As such, DEOH-XGBoost effectively uncovers the intrinsic connections between multi-type data to achieve accurate carbon emission prediction. Extensive experiments demonstrate that our DEOH-XGBoost has significantly better prediction accuracy than related state-of-the-art methods. Our source code and datasets can be found at the following link: https://github.com/lingfei0804/DEOHXGBOOST.