An Enhanced Fault Prediction Model for Embedded Software based on Code Churn, Complexity Metrics, and Static Analysis Results

摘要

Software systems evolve over time because of functionality extensions, changes in requirements, optimization of code, fixes for security and reliability bugs, etc., and it is commonly known that software quality assurance is thus a continuous issue and is often extremely time-consuming. Therefore, techniques to obtain early estimates of fault-proneness can help in increasing the efficiency and effectiveness of software quality assurance. The ability to predict which components in a large software system are most likely to contain the largest numbers of faults in the next release helps to better manage projects, including early estimation of possible release delays, and affordably guide corrective actions to the quality of the software. This paper extends our previous work, where we demonstrated that the combination of code complexity metrics together with static analysis results allows accurate prediction of fault density and to build classifiers discriminating faulty from non-faulty components. The extension presented in this paper augments our predictor and classifier with code churn metrics. We applied our methodology to C++ projects from Daimler's head unit development. In experiments to separate fault-prone from non-fault-prone components, our new approach achieved a classification accuracy of 89%, and the regressor predicted the fault density with an accuracy of 85.7%. This is an improvement of 7.5% with respect to the accuracy of fault density prediction, and an improvement of 10% to the accuracy of fault classification compared to our previous approach that did not take code churn metrics into account.