Abstract:According to the characteristics of the software failure data, the Least Square Support Vector Regression Machine (LSSVRM) was applied to build the software failure model. The LSSVRM was used to construct a fitting model according to the failure data that is a small sample and the parameters in LSSVRM were optimized by the simulated annealing algorithm to get the SA-LSSVRM and further improve the fitting model. Compared with the failure mode of NonHomogeneous Poisson Process (NHPP), the failure model based on the LSSVRM and SA algorithm has the higher fitting precision and the adjustments of the model parameters could be simplified.