An accurate assessment of the utilization of a catalyst for fuel cells is beneficial for guiding the optimization of working conditions and adjustment of parameters. In this study, a composite model containing the numerical model of catalyst layers and improved support vector machine (SVM) is proposed to predict the catalyst utilization. The effectiveness factor of the field of catalytic reaction engineering is introduced into the numerical model. The adaptive learning factor and the differential evolution strategy are introduced to improve the ability of generalization and robustness of the distinguishing method in the SVM. The results indicate that the radial basis function is well-suited as the kernel function of the SVM, and the accuracy of the prediction in the anode and cathode can be increased up to 99.31% and 99.65%, respectively. The increasing transfer coefficient leads to an enhancement of the catalytic reaction, and the particle size of the catalyst impacts the catalytic reaction mainly by changing the pattern of internal diffusion. Pressure has little impact on the Knudsen diffusion.