Can genetic programming improve software effort estimation? A comparative evaluation

Accurate software effort estimation is an important part of the software pr ocess. Originally, estimation was performed using only human expertise, but more recently, attention has turned to a variety of machine learning (ML) methods. This paper attempts to evaluate critically the potential of geneti c programming (GP) in software effort estimation when compared with previou sly published approaches, in terms of accuracy and ease of use. The compari son is based on the well-known Desharnais data set of 81 software projects derived from a Canadian software house in the late 1980s. The input variabl es are restricted to those available from the specification stage and signi ficant effort is put into the GP and all of the other solution strategies t o offer a realistic and fair comparison. There is evidence that GP can offe r significant improvements in accuracy but this depends on the measure and interpretation of accuracy used. GP has the potential to be a valid additio nal tool for software effort estimation but set up and running effort is hi gh and interpretation difficult, as it is for any complex meta-heuristic te chnique. (C) 2001 Elsevier Science B.V. All rights reserved.