A genetic programming approach to software cost modeling and estimation

This paper investigates the utilization of Genetic Programming (GP) as a method to facilitate better software cost modeling and estimation. The aim is to produce and examine candidate solutions in the form of representations that utilize operators and operands, which are then used in algorithmic cost estimation. These solutions essentially constitute regression equations of software cost factors, used to effectively estimate the dependent variable, that is, the effort spent for developing software projects. The GP application generates representative rules through which the usefulness of various project characteristics as explanatory variables, and ultimately as predictors of development effort is investigated. The experiments conducted are based on two publicly available empirical datasets typically used in software cost estimation and indicate that the proposed approach provides consistent and successful results