Validating and understanding software cost estimation models based on neural networks

Abstract EN

Software development effort estimation with the aid of artificial neural networks (ANN) attracted considerable research interest at the beginning of the nineties.

However, the lack of a natural interpretation of their estimation process has prevented them from being accepted as common practice in cost estimation.

Indeed, they have generally been viewed with skepticism by a majority of the software cost estimation community.

In this paper, we investigate the use and the interpretation of the Radial Basis Function Networks (RBFN) in the software cost estimation field.

We first apply the RBFN to estimating the costs of software projects, and then study the interpretation of cost estimation models based on an RBFN using a method which maps this neural network to a fuzzy rule-based system, taking the view that, if the fuzzy rules obtained are easily interpreted, then the RBFN will also be easy to interpret.

Our case study is based on the COCOMO’81 dataset.