Abstract: We propose a two-stage combination prediction model called GSPS-back propagation neural network (BPNN) based on support vector machine, genetic algorithm, particle swarm optimization and BPNN. The first stage of the model provides more features to handle data effectively for the second stage, reduce the data dimensions, and improve the convergence of the algorithm. The second stage of the model addresses the overfitting issue that may occur in the first stage. Two numerical examples are presented. Results show that the GSPS-BPNN model has better accuracy and stability than a one-stage prediction model.