Abstract: Based on principal component analysis (PCA) and improved N-W nonparametric estimation method (INW),a new software failure prediction model was presented.First of all,through the principal component analysis of training sample set of nonparametric estimation,the input number of nonparametric method was reduced.Then the variancecontribution ratio of PCA was used as the weight of the bandwidth matrix in nonparametric estimation method,the impact of each imput factor on the results was eliminated in a different extent and software failure prediction models were built.Finally,this paper gave example analysis based on one real software failure data set Eclipse JDT.The results show that the failure prediction model based on improved nonparametric method has made further improvement in prediction precision and stability. Within the forecast range of the last ten steps,the average error of predictive value is 16.2575,and the mean square error is 0.0726.

Key words: Software failure,Principal component analysis (PCA),N-W nonparametric estimation,Bandwidth