Intelligent data-driven models for the accurate multi-factor prediction of carbon credit prices

Accurately pricing carbon credits is essential for maintaining a transparent and effective carbon market. However, carbon credit price time series data are non-stationary, nonlinear, and multicollinear. This study addresses these challenges by developing advanced multi-factor prediction models for short-term price forecasting. The proposed models integrate key factor identification and optimised prediction algorithms to manage complex interactions among 22 external factors. This paper proposes a carbon credit multi-factor identification (CCMFI) model to study the importance of each factor. The CCMFI model combines the random forest (RF) model and SHapley Additive exPlanations (SHAP). The selected key factors, were then used as input to the carbon credit multi-factor prediction (CCMFP) models. The CCMFP applies feature extraction and reduction techniques, including independent component analysis (ICA), nonlinear independent component analysis (NLICA), and principal component analysis (PCA). The extracted components then serve as input variables for the support vector regression (SVR) and multilayer perceptron (MLP) neural networks. The study used real daily price data for Australian Carbon Credit Units (ACCUs) to validate the proposed model. The experiment results demonstrate that the ICA_SVR model with $$K = 10$$ factors outperformed other models, achieving an MSE value of 1.039, a $$R^2$$ value of 0.99, and a computational efficiency of 1.33 s. All proposed models exhibited superior prediction performance, with accuracy exceeding 97%. The developed models improve confidence among carbon credit traders and investors, helping mitigate the financial risks associated with price fluctuations. This research supports global efforts by promoting carbon credits as an effective tool for sustainable practices.