Alternatives to regression models for estimating software projects
Stephen G. MacDonell, Andrew Robert Gray · Tuwhera (Auckland University of Technology) · 1996
The use of `standard' regression analysis to derive predictive equations for software development has recently been complemented by increasing numbers of analyses using less common methods, such as neural networks, fuzzy logic models, and regression trees. This paper considers the implications of using these methods and provides some recommendations as to when they may be appropriate. A comparison of techniques is also made in terms of their modelling capabilities with specific reference to function point analysis. 1 Introduction Effective means of project effort estimation have been sought since the advent of the area of research and practice now commonly referred to as software metrics. Halstead, one of the founders of software measurement, included in his inspired (but subsequently seen as somewhat flawed) assertion of software science an equation to predict program development effort based on fundamental algorithm size (Halstead 1977). As understanding of the software process has ...