Software defect prediction approaches

For the software being developed, it is natural to have some degree of defects. What is important is to be able to find these defects. Testing activities are essential but testing each fragment of the software is impossible and defects still occur even after several detailed test activities. Therefore, there is a need for effective methods to detect bugs in software. It is well known that it's less expensive to find software defects early on in program. Research shows that finding and fixing a software problem after delivery is often 100 times more expensive than the requirements and design phase. About 80% of the defects come from 20% of the modules, and about half the modules are defect free. Considering these facts, if fault-prone modules are detected, software test efforts can be focused on these modules. By using the limited resources we have on the modules having the highest possibility to have defects, defect detection rate will increase and it will be possible to gain valuable resources by finding defects early. Serving this purpose, activities named as detection of fault-prone modules or defect prediction, help to detect the presence of defects as early as possible in an automated fashion. These approaches differ according to their techniques and success rates. The main approaches in the field will be inspected and an effective model is aimed to be developed in order to predict software entities having bugs. In our study, an approach based on logistic regression analysis and focusing on static software code metrics is utilized. A public bug database and an ATM monitoring software source code which is developed by Provus Bilişim Hizmetleri A.Ş., are used for the creation of the model and to find the performance of the study.