Software Reliability Model Selection Based on Deep Learning
Yoshinobu Tamura, Mitsuho Matsumoto, Shigeru Yamada · 2016
In the past, many software reliability models have been proposed by several researchers. Several model selection criteria such as Akaike's information criterion, mean square errors, predicted relative error and so on, are used for the selection of optimal software reliability models. These assessment criteria can be useful for the software managers to assess the past trend of fault data. However, it is very important to assess the prediction accuracy of model after the end of fault data in the actual software project. In this paper, we propose a method of optimal software reliability model selection based on the deep learning. Also, we show several numerical examples of software reliability assessment in the actual software projects. Moreover, we compare the methods to estimate the cumulative numbers of detected faults based on the deep learning by using the fault data sets of actual software projects.