Ensemble Software Development Effort Estimation Using Data Envelopment Analysis
Konstantinos Charmanas, Nikolaos Mittas, Lefteris Angelis · 2020
The field of Software Development Software Estimation in Software Engineering is critical due to its practical importance and very challenging due to the lack of a globally best model, able to predict accurately the effort (and therefore the cost) of any new software project. After a long research period on different models and their improvements, the research interest is directed towards ensemble methods, i.e. methods which combine the results of different single models. Furthermore, it is desirable to characterise the accuracy of models by different criteria. In this study, we develop a methodology based on the Data Envelopment Analysis technique, well-known in operation research, so as to rank different models based on multiple criteria and then to combine the best of them in order to achieve better prediction performance. The experimentation involves 93 models applied to 10 datasets and provides very promising results regarding the performance of the proposed ensemble approach.