Development and improvement of statistical methods supporting quality software processes

我们开发了支持质量软件构建过程的统计方法。该方法可以将软件开发的上层过程中获得的有用信息引入到基于测试阶段的软件可靠性评估中，测试阶段是软件开发的最后一个过程。这有助于提高可靠性评估的准确性。在文献中，它是已知的，多元回归模型可以用来处理这样的多元信息从上层的过程，但从我们的结果，SEM（结构方程模型），GMDH（分组方法的数据处理），和神经网络的方法可以显示更好的可靠性评估的基础上的真实的数据分析。特别是，神经网络方法是我们调查中最好的方法。提出了一种新的Bootstrap软件可靠性评估方法，并在软件可靠性评估工具SafeMan上实现。

We developed statistical methods supporting the quality software construction processes. Our methods can introduce useful information obtained from the upper process of software development into software reliability assessment based on the testing phase, which is the last process of the software development. This helps to improve the accuracy of the reliability assessment. In the literature, it is known that the multiple regression models can be used to treat such multivariate information from the upper process, but from our results, SEM (structure equation models), GMDH (group method of data handling), and neural network approach could show the better reliability assessment based on the real data analysis. In particular, the neural network approach is the best method in our investigation. Also, a new bootstrap method for software reliability assessment was proposed and implemented to the software reliability tool SafeMan.