In this paper, a modified version of the support vector machine (SVM) is proposed as an empirical model of polymerization processes. Polymerization processes are highly nonlinear and have a large number of input variables; hence, some qualities of their products must be estimated using an inference model rather than a principle model. The proposed method is derived by modifying the risk function of the standard SVM with the use of locally weighted regression. This method treats the correlations among the many process variables and nonlinearities, using the concept of smoothness. The case studies show that the proposed method exhibits superior performance to that of standard support vector regression, which is itself superior to the traditional statistical learning machine, in regard to treating high-dimensional, sparse, and nonlinear data.