Approach for Test Profile Optimization in Dynamic Random Testing

Dynamic Random Testing (DRT) is a feedback-based software testing strategy, which has been proved to be more effective than the traditional Random Testing (RT) and Random-Partition Testing (RPT) strategies. The major advantage of DRT is that the test profile is dynamically adjusted based on the previous test data. Since the frequency and range of the profile adjustment are fixed during the testing process, DRT might not react in time to the changes of the defect detection rates. In order to overcome these shortcomings, an approach for the test profile optimization in DRT, denoted as O-DRT, is proposed in this paper. In O-DRT, the test profile adjustment contains two parts. In addition to the original adjustment in DRT, O-DRT will change the test profile to a theoretically optimal one when the pre-defined criterion is satisfied. The theoretically optimal test profile is calculated based on an optimization goal of both maximizing the overall defect detection rate and minimizing its variance. Experiments on five real-life software subjects are conducted to validate the effectiveness of O-DRT. Experimental results demonstrate that O-DRT outperforms RPT and DRT in terms of the number of test cases required to detect and remove a given number of defects.