Effort estimation in software development is acknowledged as one of the key success factors of systems creation. Despite significant interest in the subject and multiple existing methods, effort estimation remains a challenge with low prediction accuracy and common effort overruns. The authors attempt to improve the performance of the analogy-based estimation (ABE) method. ABE uses attributes of previously completed similar projects (such as size, type, programming language, and development platform) to estimate efforts for future projects.
The suggested model uses two steps. The first step is clustering/classifying projects in the historical dataset into groups by analysis of their attributes. Classification may employ c-means or k-means clustering. Next, the new project is compared with historical projects with similar key attributes. The experiment has been performed on three real datasets: International Software Benchmarking Standard Group (ISBSG 2011), COCOMO 81, and Maxwell.
The authors conclude that the new method provides promising results compared to the standard ABE method. However, they admit that the use of the new method can be threatened by the absence of a commonly accepted approach to the selection of the key attributes and a lack of historical data attributes. The use of the mean magnitude of relative error (MMRE) and the percentage of the prediction (PRED) as accuracy measures has generated certain debate regarding their efficacy, but this issue is not addressed by the authors.
Academics and practitioners working on software estimation topics are the paper’s intended audience. Readers interested in this topic can find additional information on the subject in other sources [1,2,3].