A regression forecasting model of carbon dioxide concentrations based-on principal component analysis-support vector machine

We propose Principal Component Analysis-Support Vector Machine (PCA-SVM) to forecast the changes of regional carbon dioxide concentrations. Firstly, we get the most valuable principal components of the influencing factors (IF) of carbon dioxide concentrations by PCA. Then we use the output of PCA as the input of non-linear SVM to learn a regression forecasting model with radial basis function. Due to the introducing of PCA, we successfully eliminate the redundant and correlate information in IFs and reduce the computation cost of SVM. The results of the comparative experiment demonstrate that our PCA-SVM model is more effective and more efficient than the standard SVM. Moreover, we have tested different kernel functions in our PCA-SVM model, and the experimental results show that PCA-SVM model with radial basis function performs best respecting to the learning ability and generalization capability.