Identification method for defect-introducing fine-grained software changes

Software defects were introduced into software system by software changes in the software development process. A new method to identify defect-introducing fine-grained changes was proposed to improve the efficiency of defect finding and reduce the cost of manual inspection. This method was based on the idea of machine learning classification. It took the fine-grained change as classification instance and constructed feature set from five dimensions, namely time, context, content, purpose and implementer of the change. It built fine-grained change instances automatically by mining software history repositories with the program static analysis and natural language semantic analysis techniques. It trained a classifier by learning change instances in software history, which could identify whether a new fine-grained change introduced any defects or not. Cost-effectiveness analysis was conducted on real software systems to verify the validity of the proposed method. The results indicate that compared with methods for file and transaction level changes, this method can reduce the manual inspection cost significantly.