A replicated study comparing web effort estimation techniques

The objective of this paper is to replicate two previous studies that compared at least three techniques for Web effort estimation in order to identify the one that provides best prediction accuracy. We employed the three effort estimation techniques that were mutual to the two studies being replicated, namely Forward Stepwise Regression (SWR), Case-Based Reasoning (CBR) and Classification & Regression Trees (CART). We used a cross-company data set of 150 Web projects from the Tukutuku data set. This is the first time such large number of Web projects is used to compare effort estimation techniques. Results showed that all techniques presented similar predictions, and these predictions were significantly better than those using the mean effort. Thus, all the techniques can be exploited for effort estimation in the Web domain, also using a cross-company data set that is specially useful when companies do not have their own data on past projects from which to obtain their estimates, or that have data on projects developed in different application domains and/or technologies. © Springer-Verlag Berlin Heidelberg 2007