Assessing Single and Ensemble KNN Methods for Estimating Software Testing Effort

Accurate estimation of effort in software development projects is crucial for their successful execution but poses significant challenges, particularly during the testing phase. Software testing directly impacts the quality and reliability of the final product. Traditional methods for estimating testing effort-typically assigning a fixed percentage (10% a 60%) of the total projected Software Development Life Cycle effort-can be risky as they may not adequately account for the specific complexities involved. This can lead to suboptimal resource allocation and potential project delays. Despite the critical need for precise testing effort estimation, there is a notable lack of research dedicated to developing accurate and systematic methods for this phase. This paper aims to address this gap by employing K-Nearest Neighbors (KNN) to estimate the effort required for software testing. Various KNN variants were examined, incorporating six different distance metrics: Minkowski, Cityblock, Cosine, Euclidean, Mahalanobis, and Chebyshev. Both single and ensemble KNN methods were evaluated using the ISBSG dataset. The results revealed that while certain KNN variants, particularly those using the Cityblock distance metric, performed well, there was no significant difference between single and ensemble methods. These findings underscore the potential of KNN in improving effort estimation practices for software testing, although the advantage of ensemble approaches over single models was not statistically significant.