The mechanism of the burning side reaction in the INVISTA oxidation process is complex, nearly unknown, and difficult to model. In this study, a radial basis function neural network (RBFNN) was used to model the burning side reaction based on the data collected from the INVISTA process. Over the past decades, clustering methods have been used to improve the ability of several RBFNN models to determine more efficient structures. However, these RBFNN models determine the RBFNN structure without considering the prediction accuracy of the model. To elucidate the optimal RBFNN structure and obtain a satisfactory burning side reaction model, RBFNN integrated with partial mutual information-least square regression (PMI-LSR) is proposed. PMI-based selection takes the correlation between the hidden layer and the output layer into account and eliminates redundant information in the selected hidden layer neurons to improve RBFNN prediction accuracy. Sammon's nonlinear map is used to illustrate the distribution of the selected hidden layer centers. This distribution differs from the uniform distribution of the cluster centers obtained using cluster methods. The burning side reaction model developed by PMI-LSR-RBFNN is better than those obtained by several cluster based RBFNN variants.