Abstract: In view of the traditional software defect prediction model, the problem of low accuracy and poor adaptability, this paper proposes a new software prediction model (IVPSO-BP) with better performance and effect, and combining with two by improving PSO algorithm for optimizing BP. Firstly, To improve Particle Swarm Optimization, and use it to optimize BP. Secondly, this paper employs optimized BP to build a prediction model. Finally, compares the experiment results with other machine learning methods-BP, J48 and PSO-BP. Through analyzing the data of the final experiment, the results indicated that proposed method owe a higher prediction precision.