An Experiment of Different Similarity Measures on Test Case Prioritization for Software Product Lines

Abstract

Software product line (SPL) engineering paradigm is commonly used to manage variability and commonalities of business applications to satisfy a specific need or goal of a particular market. However, due to time and space complexity, combinatorial interaction testing (CIT) has been suggested to reduce the size of test suites. Although CIT is known as a promising approach to overcome these problems, there are still issues such as combinatorial explosion of features, which drains budget allocated for testing. Therefore, test case prioritization (TCP) is preferred to gain a better result in terms of producing an efficient detection of faults. Among prioritization techniques used in regression testing is similarity-based test case prioritization. Similarity-based test case prioritization rearranges test cases through calculation of distance between test cases using similarity measures. Result from the use of similarity measures in test case prioritization contributes to a much better testing process. This paper provides a comparison of selected similarity measures to investigate the feasibility and suitability of similarity measures to be used in SPL through experimentation. Jaccard, Hamming, Jaro-Winkler, Cosine similarity, Counting, and Sorensein distances have been chosen as similarity measures in this study. The result showed JaroWinkler as the best similarity measure with an 84.96% Average Percentage of Faults Detected (APFD) value across eight feature models. The study offers insights on similarity measures in SPL context. Further, the paper concludes with suggestions on room for improvement, which could be achieved through experimentation and comparison studies.