562015-03-24 , 法政大学大学院理工学・工学研究科
Software reliability is more enhanced if a software development project manager can predict whether the number of failures after the software release is zero or not, during the software development process. In response to this fact, we started this investigation which predicts the software reliability by using the multivariate analysis approach. Unfortunately, the objective data set we obtained has a lot of missing values. Therefore in this study, we first prepared two data sets. One is the complete data set extracted from the raw data set, and the other was created as a virtual complete data set by means of the data imputation method so as to be similar to the complete data set. As a result of the software reliability prediction, we have found that our imputation method is comparable in terms of the prediction accuracy between two data sets via the discriminant analysis. Therefore it is shown that the data sets with our missing value treatment method can be used for software reliability prediction as well as the complete data sets.