Adaptive random testing with filtering: An overhead reduction technique

Adaptive Random Testing (ART) is an approach to testing software based on Random Testing (RT), but incorporating additional mechanisms to ensure a more widespread and even distribution of test cases over the input domain, it has been found that ART, under certain conditions, can significantly outperform RT, in terms of number of test cases required to detect a failure (a measure referred to as the F-measure). One implementation of ART, based on the use of exclusion zones and restriction of test case selection to outside of these zones, is Restricted Random Testing (RRT). In this paper, we present an overview of the basic RRT method, using circular and spherical exclusion regions, and then introduce an alternative exclusion shape, motivated by the promise of lower computational costs. Investigation into this alternative shape (square) exclusion method lead to a hybrid implementation of RRT. called filtering. Filtering enables the combination of the computationally cheaper square exclusion shape and the faster (for failure finding) original, circular exclusion shape. Simulation and experimental evidence are also presented supporting the methods.