Applying software metrics to RNN for early reliability evaluation

Abstract

Structural modeling is an important branch of software reliability modeling. It works in the early reliability engineering to optimize the architecture design and guide the later testing. Compared with traditional models using test data, structural models are often difficult to be applied due to lack of actual data. A software metrics-based method is presented here for empirical studies. The recurrent neural network (RNN) is used to process the metric data to identify defeat-prone code blocks, and a specified aggregation scheme is used to calculate the module reliability. Based on this, a framework is proposed to evaluate overall reliability for actual projects, in which algebraic tools are introduced to build the structural reliability model automatically and accurately. Studies in two open-source projects show that early evaluation results based on this framework are effective and the related methods have good applicability.