Towards a Reliability Prediction Model based on Internal Structure and Post-Release Defects Using Neural Networks

Abstract

Reliability is one of the most important quality attributes of a software system, addressing the system’s ability to perform the required functionalities under stated conditions, for a stated period of time. Nowadays, a system failure could threaten the safety of human life. Thus, assessing reliability became one of the software engineering‘s holy grails. Our approach wants to establish based on what project’s characteristics we obtain the best bug-oriented reliability prediction model. The pillars on which we base our approach are the metric introduced to estimate one aspect of reliability using bugs, and the Chidamber and Kemerer (CK) metrics to assess reliability in the early stages of development. The methodology used for prediction is a feed-forward neural network with back-propagation learning. Five different projects are used to validate the proposed approach for reliability prediction. The results indicate that CK metrics are promising in predicting reliability using a neural network model. The experiments also analyze if the type of project used in the development of the prediction model influences the quality of the prediction. As a result of the operated experiments using both within-project and cross-project validation, the best prediction model was obtained using PDE (PlugIn characteristic) for MY project (Task characteristic).