Unsupervised Machine Learning Approaches for Test Suite Reduction

Abstract

ABSTRACT Ensuring quality and reliability mandates thorough software testing at every stage of the development cycle. As software systems grow in size, complexity, and functionality, the parallel expansion of the test suite leads to an inefficient utilization of computational power and time, presenting challenges to optimization. Therefore, the Test Suite Reduction (TSR) process is of great importance, contributing to the reduction of time and costs in executing test suites for complex software by minimizing the number of test cases to be executed. Over the past decade, machine learning-based solutions have emerged, demonstrating remarkable effectiveness and efficiency. Recent studies have delved into the application of Machine Learning (ML) in the software testing domain, where the high cost and time consumption associated with data annotation have prompted the use of unsupervised algorithms. In this research, we conducted a Systematic Mapping Study (SMS), examining the types of unsupervised algorithms implemented in developed models and thoroughly exploring the evaluation metrics employed. This study highlighted the prevalence of the K-Means clustering algorithm and the coverage metric for validation in various studies. Additionally, we identified a gap in the literature regarding scalability considerations. Our findings underscore the effective use of unsupervised learning approaches in test suite reduction.