Abstract:
In the world of software development, organizations must optimize the use of their limited resources to deliver quality products on time and within budget. This requires efficient and effective discovery, removal, and prevention of faults introduced during the development process or residual faults from maintenance stage. To reveal software fault, testing is generally employed by procedurally running the system with adequate test cases. Such an undertaking usually incurs high costs and considerable efforts. This dissertation proposes an approach for software fault prediction and fault location without actually running the software. The process of software fault prediction consists of four stages, namely, fault-prone prediction, fault type prediction, dynamic fault prediction, and fault locating. Fault predictive models are constructed based on software metrics with the help of fuzzy logic and neural network techniques for each stage. Once identified, all potential faults are pinpointed to locate their whereabouts. The results are further analyzed to obtain principal metrics that are conducive to fault prediction with the help of sensitivity analysis process. Hence, the proposed approach will furnish a basis for machine learning building blocks that could be realized in software quality assurance, whereby replacing time-consuming and error-prone inspection process to attain more reliable software products.
