Automation and Evaluation of Software Fault Prediction

Abstract: Delivering a fault-free software to the client requires exhaustive testing, which in today's ever-growing software systems, can be expensive and often impossible. Software fault prediction aims to improve software quality while reducing the testing effort by identifying fault-prone modules in the early stages of development process. However, software fault prediction activities are yet to be implemented in the daily work routine of practitioners as a result of a lack of automation of this process. This thesis presents an Eclipse plug-in as a fault prediction automation tool that can predict fault-prone modules using two prediction methods, Naive Bayes and Logistic Regression, while also reflecting on the performance of these prediction methods compared to each other. Evaluating the prediction methods on open source projects concluded that Logistic Regression performed better than Naive Bayes.As part of the prediction process, this thesis also reflects on the easiest metrics to automatically gather for fault prediction concluding that LOC, McCabe Complexity and CK suite metrics are the easiest to automatically gather.