Risky Module Estimation in Safety-Critical Software

Software used in safety-critical system must have high dependability. Software testing and V&V (Verification and Validation) activities are very important for assuring high software quality. If we can predict the risky modules in safety-critical software, testing activities and regulation activities can be applied to them more intensively. In this paper, we classify the estimated risk classes which can be used for deep testing and V&V. We predict the risk class for each module using support vector machines. We can consider that the modules classified to risk class 5 or 4 are more risky than others relatively. For all classification error rates, we expect that the results can be useful and practical for software testing, V&V, and activities for regulatory reviews. In the future works, to improve the practicality, we will have to investigate other machine learning algorithms and datasets.