Adaptive Random Testing by Bisection and Comprehensive Distance

Adaptive random testing (ART) has been proved to be effective in improving the failure detection ability of random testing. As a lightweight ART algorithm, the ART by bisection (ART-B) can realize test case generation in a linear order of time complexity, but its ability for finding failures is not so strong. In this study, the dynamic bisection is used to generate candidates randomly from the empty regions as much as possible. For each candidate, two types of distances are taken into account. Then, a comprehensive distance metric is defined to determine the next test case from the candidate set. For the nearest neighbor query in the proposed ART by Bisection and Comprehensive Distance (ART-BCD), the distance-ware forgetting is adopted to ensure its computational cost is still in the linear order. To validate the effectiveness of ART-BCD algorithm, both simulation experiments and empirical studies are performed for comparative analysis. The experimental results show that ART-BCD is better than or comparable to the ART-B and other typical algorithms, such as the fixed-size-candidate-set ART (FSCS-ART), in most cases, especially for the block failure pattern.