Fault Proness Estimation of Software Modules Using Machine learning

To evaluate the software quality it is required to measure factors affecting the qualityof the software . Reliability and number of faults are examples of quality factors. If these factors are measured during the software development life cycle, more efficient and optimal activities can be performed to improve the software quality. The difficulty is that these factors could be measured only at the ending steps of the life cycle. To resolve the difficulty, these factors are indirectly measured through some software metrics which are available at the early stages of life cycle. These metrics are used as the input to fault prediction models and software components which may be faulty are the output of these models. Prediction of fault prone modules is a well known approach in software testing phase. When a module is predicted to be faulty, apparently more efforts have to be paid for correcting it. In addition to the module, all its dependent modules require specific consideration. When modifying a module all its dependent modules may be affected. The difficulty is that current known metrics for fault prediction do not reflect this situation. To resolve the difficulty, in this thesis, new metrics are introduced. Our experimental results show that the more the dependees of a module are changed, the more fault prone the module will be