Dynamic random testing with test case clustering and distance-based parameter adjustment

Abstract Context Software testing is essential in software engineering to improve software reliability. One goal of software testing strategies is to detect faults faster. Dynamic Random Testing (DRT) strategy uses the testing results to guide the selection of test cases, which has shown to be effective in the fault detection process. Objective Previous studies have demonstrated that DRT is greatly affected by the test case classification and the process of adjusting the testing profile. In this paper, we propose Distance-based DRT (D-DRT) strategies, aiming at enhancing the fault detection effectiveness of DRT. Method D-DRT strategies utilize distance information of inputs into the test case classification and the testing profile adjustment process. The test cases are vectorized based on the input parameters and classified into disjoint subdomains through certain clustering methods. And the distance information of subdomains, along with testing results, are used to adjust the testing profile, such that test cases that are closer to failure-causing subdomains are more likely to be selected. Results We conduct empirical studies to evaluate the performance of the proposed algorithms using 12 versions of 4 open-source programs. The experimental results show that, compared with Random Testing (RT), Random Partition Testing (RPT), DRT and Adaptive Testing (AT), our strategies achieve greater fault detection effectiveness with a low computational cost. Moreover, the distance-based testing profile adjustment method is the dominant factor in the improvement of the D-DRT strategy. Conclusion D-DRT strategies are effective testing strategies, and the distance-based testing profile adjustment method plays a crucial role.