Prediction of defect proneness in web applications by domain-specific metrics

Invention of the computer was for scientific and military purposes. As the size of computers began to shrink with the improvement in the technology, their use became widespread and they started to be used for different fields, mainly commercial applications. Different programming languages have begun to emerge. With the introduction of the concept of software engineering to manage complexity in software projects, different methods have begun to be developed to facilitate the software development process. Software defect prediction is one of these methods. Each software product has to pass through the testing process before the product is released. However, as in all stages of the software development process, it is necessary to plan for the resources to be used effectively in the test phase. In order to properly plan the testing process, it is necessary to determine the sections to which the resources are to be transferred. Software defect proneness prediction techniques are used by project managers to efficiently allocate their precious resources in testing phase. These techniques give software testers some insight to decide how to create and organise test cases. Accurate prediction of defect prone modules can reduce the cost of software testing and allow project managers to act more confidently in assigning their resources. Predicting defect prone modules in software projects has been one of the most active research topics in software engineering in the last decade, and in recent years studies have shown that prediction models reached a ceiling effect. In order to improve the performance of the prediction models, the general consensus is that focus should be based on improving the datasets rather than algorithms used. We have noticed that, to this date, researchers have not benefited from domain-specific data. From this point, we choosed the web applications domain and extracted custom metric sets from some open source web applications. As a result of our experiments, we found that domain specific metrics helped us to improve the performance of software defect proneness prediction models.