Can Expert Opinion Improve Effort Predictions When Exploiting Cross-Company Datasets? - A Case Study in a Small/Medium Company

摘要

Many studies have shown that the accuracy of the predictions obtained by estimation models built considering data collected by other companies (cross-company models) can be significantly worse than those of estimation models built employing a dataset collected by the single company (within-company models). This is due to the different characteristics among cross-company and within-company datasets. In this paper, we propose an approach based on the opinion of the experts that could help in the context of small/medium company that do not have data available from past developed projects. In particular, experts are in charge of selecting data from public cross-company datasets looking at the information about employed software development process and software technologies. The proposed strategy is based on the use of a Delphi approach to reach consensus among experts. To assess the strategy, we performed an empirical study considering a dataset from the PROMISE repository that includes information on the functional size expressed in terms of COSMIC for building the cross-company estimation model. We selected this dataset since COSMIC is the method used to size the applications by the company that provided the within-company dataset employed as test set to assess the accuracy of the built cross-company model. We compared the accuracy of the obtained predictions with those of the cross-company model built without selecting the observations. The results are promising since the effort predictions obtained with the proposed strategy are significantly better than those obtained with the model built on the whole cross-company dataset.