An Approach for Choosing the Best Covering Array Constructor to Use

Covering arrays have been extensively used for software testing. Therefore, many covering array constructors have been developed. However, each constructor comes with its own pros and cons. That is, the best constructor to use typically depends on the specific application scenario at hand. To improve both the efficiency and effectiveness of covering arrays, we, in this work, present a classification-based approach to predict the "best" covering array constructor to use for a given configuration space model, coverage strength, and optimization criterion, i.e., minimizing the construction time or the covering array size. We also empirically evaluate the proposed approach by using a relatively small, yet quite realistic space of application scenarios. The approach predicted the best constructors for reducing the construction times with an accuracy of 86% and the best constructors for reducing the covering array sizes with an accuracy 90%. When two predictions were made, rather than one, the accuracy of correctly predicting the best constructors increased to 94% and 98%, respectively.