The high complexity of software is the major contributing factor of software reliability problems, and traditional parametric models may exhibit different predictive capabilities among different software projects, it is hard to select a suitable model for every software projects. Compared to traditional models, kernel based models could achieve better prediction accuracy, and had arouse the interesting of many researchers. The RVM learning scheme was applied to model the failure time data so as to capture the inner correlation between software failure time data and the m nearest failure time data. In addition, the trend of predictive accuracy with the varying of m was detected by way of Mann-Kendall test method. Thereupon, the reasonable value range of m was achieved,thus m∈｛6,7,8,9,10｝ through paired T-test in 5 common used software failure data.
