An effort prediction interval approach based on the empirical distribution of previous estimation accuracy

When estimating software development effort, it may be useful to describe the uncertainty of the estimate through an effort prediction interval (PI). An effort PI consists of a minimum and a maximum effort value and a confidence level. We introduce and evaluate a software development effort PI approach that is based on the assumption that the estimation accuracy of earlier software projects predicts the effort PIs of new projects. First, we demonstrate the applicability and different variants of the approach on a data set of 145 software development tasks. Then, we experimentally compare the performance of one variant of the approach with human (software professionals') judgment and regression analysis-based effort PIs on a data set of 15 development tasks. Finally, based on the experiment and analytical considerations, we discuss when to base effort PIs on human judgment, regression analysis, or our approach. (C) 2002 Elsevier Science B.V. All rights reserved.