An enhanced hybrid model based on multiple influencing factors and divide-conquer strategy for carbon price prediction

•A multifactor-based decomposition-integration framework for carbon price forecasting.•Develop a new two-stage data preprocessing technique to improve data input quality.•The optimized deep learning model is used to predict components with different time period features.•The proposed model shows better prediction accuracy and robustness than the baseline models. Accurate prediction of carbon price is of great significance for promoting the steady development of carbon trading market. However, as the unstable and nonlinear characteristics of carbon prices, it is challenging to obtain accurate future transaction prices. Therefore, this paper proposes a hybrid system for carbon price prediction. Specifically, the two-stage data preprocessing can decompose and reconstruct the carbon price sequence, which can reduce noise and chaotic disturbance and improve the quality of input data. Moreover, extreme gradient boosting (XGBoost) and partial autocorrelation function (PACF) are proposed to screen the influencing factors and carbon price subsequences to obtain the optimal features. The prediction and combination module uses the strategy of subsequence prediction and nonlinear integration, leveraging the strengths of each model to achieve the ideal results. In this system, an effective slime mold algorithm (SMA) is used to ensure the accuracy and stability of the prediction system. The research results show that: (a) The developed system has the best predictive performance compared to the benchmark models. (b) Energy, economy and commodity markets have a more significant long-term impact on carbon prices, and news sentiment and Baidu index contribute to their short-term fluctuations.