An Empirical Study on Software Fault Prediction Using Product and Process Metrics

Abstract

Product and process metrics are measured from the development and evolution of software. Metrics are indicators of software fault-proneness and advanced models using machine learning can be provided to the development team to select modules for further inspection. Most fault-proneness classifiers were built from product metrics. However, the inclusion of process metrics adds evolution as a factor to software quality. In this work, the authors propose a process metric measured from the evolution of software to predict fault-proneness in software models. The process metrics measures change-proneness of modules (classes and interfaces). Classifiers are trained and tested for five large open-source systems. Classifiers were built using product metrics alone and using a combination of product and the proposed process metric. The classifiers evaluation shows improvements whenever the process metrics were used. Evolution metrics are correlated with quality of software and helps in improving software quality prediction for future releases.