Using Weighted Attributes to Improve Cluster Test Selection

Abstract

Cluster Test Selection (CTS) is widely-used in observation-based testing and regression testing. CTS selects a small subset of tests to fulfill the original testing task by clustering execution profiles. In observation-based testing, CTS saves human efforts for result inspection by reducing the number of tests and finding failures as many as possible. This paper proposes a novel strategy, namely WAS (Weighted Attribute based Strategy), to improve CTS. WAS is inspired by the idea of fault localization, which ranks the program entities to find possible faulty entities. The ranking of entity is considered as a weight of attribute in WAS. And then it helps build up a more suitable distance space for CTS. As a result, a more accurate clustering is obtained to improve CTS. We conducted an experiment on three open-source programs: flex, grep and gzip. The experimental results show that WAS can outperform all existing CTS techniques in observation-based testing.