To solve the existing problems of forecasting method selection and model performance optimization in China, a method of optimizing the parameters selection for the least squares support vector machine (LS-SVM) with Improved Particle Swarm Optimization(IPSO) is proposed to carry out research on national defense project development-cost evaluation. Based on the principle of least square support vector machine, the estimation model of IPSO_LS-SVM is established by optimizing its parameter selection method, and the model training and result verification are carried out. The results show that the IPSO_LS-SVM method has higher precision and faster parameter optimization, and its estimation model is effective and advantageous.