Identifying fault prone modules: an empirical study in telecommunication system

Telecommunication systems are becoming more dependent on software-intensive products and systems. Poor software quality can threaten safety, risk a company's business, or alienate potential customers. It is no longer acceptable to ignore software quality until just prior to a product's release. This study identifies troublesome modules in a large telecommunication system. For this, we propose the software metrics of the CHILL language, which are used to develop the telecommunication software. We present the identification method of fault-prone software modules of telecommunication software using neural networks. We investigate the relationship between the proposed metrics and the change request frequency of software modules which are found during the development phase. Using the neural network model, we classify software modules as either fault prone or not fault prone based on the proposed metrics. We obtained the experimental results that the total fitting rate of 52 testing data sets was 96.2%. Therefore, for newly added software modules, we can predict whether the software module is fault prone or not.