A comparative study of hybrid models of selective classification and dynamic selection of analogies for software development effort estimation

Analogy-Based Software Development Effort Estimation (ABSDEE) approaches have been widely used by software effort researchers and practitioners. The performance of ABSDEE models can be improved by properly utilizing the characteristics of data in estimation process. This paper aims to show the effects of hybrid approaches of selective project classification and dynamic selection of number of similar historical projects on the performance of ABSDEE. In this paper we propose four hybrid ABSDEE models using selective classification and dynamic selection of analogies. The results showed that the estimation accuracy can be significantly improved with proper utilization of the inherent characteristics of a dataset and using different number of similar past projects.