An intelligently adjusted carbon price forecasting approach based on breakpoints segmentation, feature selection and adaptive machine learning

Accurate carbon price prediction is conducive to the stable operation and development of carbon financial markets. Affected by major policies and economies, the laws of carbon price may show huge changes, generating various breakpoints that hinder the prediction work. Therefore, a hybrid model based on adaptive segmentation and feature clustering is developed to forecast carbon price, which responds to the growing requirement of the high precision prediction. It solves the issue that the accuracy and stability of model are limited by the breakpoint via the adaptive processing and the fully learning for data. Meanwhile, the proposed adaptive structure improves the robustness and adaptability of prediction model, achieving accurate prediction for carbon emission trading markets with different features. The eight carbon emission trading markets in China are used to evaluate prediction performance. The obtained results indicated that the proposed model was effective and robust, with the average mean absolute error and root mean square error of only 0.2272 and 0.3321, respectively. According to the comparative analysis, the segmentation based on breakpoint and adaptive prediction based on feature clustering improve the model forecasting accuracy by 69.87% and 45.51%, respectively. Hence, the proposed model can meet the requirements of carbon financial markets and provide a benchmark for the carbon emission reduction work. [Display omitted] •An adaptive segmented prediction model is proposed to forecast the carbon price.•The segmentation is combined with feature clustering to address issue of breakpoints.•Accurate and stable prediction of carbon prices with different features is realized.•The decomposition-based structure provides benchmarks for carbon price prediction.