A comparison of modeling techniques for software development effort prediction

• Main Authors: MacDonell, SG (Author), Gray, AR (Author)
• Other Authors: Kasabov, N (Contributor), Kozma, R (Contributor), Ko, K (Contributor), OShea, R (Contributor), Coghill, G (Contributor), Gedeon, T (Contributor)
• Format: Others
• Published: Springer-Verlag, 2012-03-12T08:11:42Z.
• Subjects: Project; Conference Contribution
• Online Access: Get fulltext

 Software metrics are playing an increasingly important role in software development project management, with the need to effectively control the expensive investment of software development of paramount concern. Research examining the estimation of software development effort has been particularly extensive. In this work, regression analysis has been used almost exclusively to derive equations for predicting software process effort. This approach, whilst useful in some cases, also suffers from a number of limitations in relation to data set characteristics. In an attempt to overcome some of these problems, some recent studies have adopted less common modeling methods, such as neural networks, fuzzy logic models and case-based reasoning. In this paper some consideration is given to the use of neural networks and fuzzy models in terms of their appropriateness for the task of effort estimation. A comparison of techniques is also made with specific reference to statistical modeling and to function point analysis, a popular formal method for estimating development size and effort.