A Probabilistic Neural Network-Based Approach for Related Software Changes Detection

Current softwares are continuously updating. The change between two versions usually involves multiple program entities (e.g., Class, method, attribute) with multiple purposes (e.g., Changed requirements, bug fixing). It's hard for developers to understand which changes are made for the same purpose. However, whether two changes are related is not decided by the relationship between this two entities in the program. In this paper, we summarize 4 coupling rules (16 instances) and 4 co-changed types at class, method and attribute levels for software change. We propose the Related Change Vector (RCV) to characterize the related changes, which is defined based on the coupling rules and co-changed types. Probabilistic neural network is used to detect related software changes with RCV as input. Our approach is evaluated with experiments on 3 software projects (14 versions) written in Java. The results indicate that the average detection precision is about 90%.