In order to solve the limitation problem of using the traditional parameter method to predict the research costs, it adopts the improved particle swarm optimization (PSO) to improve the LSSVM model, which constructing the development cost’s forecasting model. It uses two kinds of optimization strategy to improve the PSO, which can control the population initialization process, and overcome the shortcomings that the particle swarm algorithm is easy to early maturity. It uses the improved particle swarm algorithm to optimize the model parameters and nuclear parameters of the least square support vector machine (LSSVM) in order to get better prediction effect. The prediction results show that the prediction model used in the cost military engineering machinery, is obviously superior to the traditional forecasting model. The improved prediction model has the very good prediction accuracy and efficiency.