Existing estimation frameworks generally provide one-size-fits-all solutions that fail to produce accurate estimates in most environments. Research has shown that the accomplishment of accurate effort estimates is a long-term process that, above all, requires the extensive collection of effort estimation data by each organization. Collected data is generally characterized by a set of attributes that are believed to affect the development effort. The attributes that most affect development effort vary widely depending on the type of product being developed and the environment in which it is being developed. Thus, any new estimation framework must offer the flexibility of customizable attribute selection. Moreover, such attributes could provide the ability to incorporate empirical evidence and expert judgment into the effort estimation framework. Finally, because software is virtual and therefore intangible, the most important software metrics are notorious for being subjective according to the experience of the estimator. Consequently, a measurement and inference system that is robust to subjectivity and uncertainty must be in place. The Effort Estimation Framework with Customizable Attribute Selection (EEF-CAS) presented in this paper has been designed with the above requirements in mind. It is accompanied with four preparation process steps that allow for any organization implementing it to establish an estimation process. This estimation process facilitates data collection, framework customization to the organization’s needs, its calibration with the organization’s data, and the capability of continual improvement. The proposed framework described in this paper was validated in a real software development organization.
Available at: http://works.bepress.com/katarina_grolinger/7/