Using a competitive learning neural network to evaluate software complexity

With recent advances in neural networks, an increasing number of application areas are being explored for this technology. Also, as software takes a more prominent role in systems engineering, ensuring the quality of software is becoming a critical issue. This paper explores the application of one neural network paradigm—the competitive learning network—to the problem of evaluating software complexity. The network was developed by ARINC Research Corporation for its SofTest software analysis system, developed on a Sun workstation. In this paper, we discuss the network used in SofTest and the approach taken to train the network. We conclude with a discussion of the implications of the approach and areas for further research.