Abstract
The manufacture of products in the industrial sector is the principal source of carbon emissions. To slow the progression of global warming and advance low-carbon economic development, it is essential to develop methods for accurately predicting carbon emissions from industrial sources and imposing reasonable controls on those emissions. We select a support vector machine to predict industrial carbon emissions from 2021 to 2040 by comparing the predictive power of the BP (backpropagation) neural network and the support vector machine. To reduce noise in the input variables for BP neural network and support vector machine models, we use a random forest technique to filter the factors affecting industrial carbon emissions. The statistical results suggest that BP's neural network is insufficiently adaptable to small sample sizes, has a relatively high error rate, and produces inconsistent predictions of industrial carbon emissions. The support vector machine produces excellent fitting results for tiny sample data, with projected values of industrial carbon dioxide emissions that are astonishingly close to the actual values. In 2030, carbon emissions from the industrial sector will have reached their maximum level.