Assessing RBFN-based software cost estimation models

This paper is concerned with the design of the neural networks approach, especially Radial Basis Function Neural (RBFN) networks, for software effort estimation models. The study firstly focuses on the construction of the RBFN middle layer composed of receptive fields, using two clustering techniques: hard C-means and fuzzy C-means. Thereafter, we evaluate and compare the performance of effort estimation models that use an RBFN construction-based either on hard or fuzzy C-means. This study uses the ISBSG dataset and confirms the usefulness of an RBFN-based on fuzzy C-means for software effort estimation. Copyright © 2013 by Knowledge Systems Institute Graduate School.