The bonding performance of the interface between two concrete layers is of great importance for China Rail Track System III (CRTS III) slab ballastless track structure. While the bonding performance is close to the distribution and amounts of defects such as the bubble or the void on the bonding interface. In this paper, combining the Support Vector Machine (SVM) with the Particle Swarm Optimization (PSO) algorithm, named SVM-PSO algorithm is used to predict the splitting tensile strength of the bonding interface based on the distribution of defects. And the influence of different parameters in the SVM-PSO algorithm on the prediction ability is discussed. Results indicate that the relative error between the average strength value of every 100 predictions and the experimental strength value is less than 5%. The achievements will provide an effective method for predicting the splitting tensile strength of the bonding interface between two concrete layers like CRTS III slab ballastless track structure in practice.