Conclusion Stability on Performance of Analogy-Based Software Effort Estimation
Passakorn Phannachitta · NAIST Digital Library (Nara Institute of Science and Technology) · 2016
Analogy-based estimation (ABE) is one of the most successful methods to estimate the required amount of effort for a new software development project.According to our literature review, the excellent estimation accuracy of ABE is strongly associated with approaches adopted in its 5 essential components: normalization of software project features, feature subset selection, similarity measures, solution adaptation, and the number of analogues.Being a very successful effort estimation method, researchers have continually proposed new approaches to these 5 essential components, combined the approaches, and tailored them to the ABE models mainly for performance improvements.To date, it has been reported that over thousands of combinations of approaches are existed; however, to the best of our knowledge, no one could successfully determine the best one.The problem persists mainly due to conflicting research conclusions frequently shown in past studies of ABE, where different studies often produced divergent performance results.On the contrary, being able to solve this problem is important for a wide range of scientific and industrial processes, that is, unless a stable conclusion on the performance can be drawn, it would be difficult for industrial practitioners to be able to access a sufficiently accurate model, resulting in increasing risk of failed software project.Also for the research community, lack of sufficiently effective models may limit the potential improvement of future model proposals, since any new approaches maybe evaluated with inadequate standard.This brings to mind that being able to determine a stable conclusion on the