Enhancing software effort estimation through influencers-based project similarity measurement

Abstract

This paper introduces a novel methodology for enhancing software effort estimation accuracy by incorporating observed ratings into measuring project similarity. Unlike traditional methods that rely only on historical project data, the proposed method leverages observed ratings to identify influencers within the dataset. These influencers serve as critical references that guide the estimation process, transforming project representations into a fully specified space where the similarity between projects can be accurately calculated. The significance of our method is that it overcomes the limitations of existing effort estimation methods by incorporating additional contextual information provided by observed ratings, thereby improving effort estimation accuracy. Experimental results on the ISBSG dataset show that our approach achieved better Root Mean Squared Error (RMSE) results than other neighbor-based effort estimation methods. Our approach offers a promising avenue for more accurate and data-driven effort estimation, enabling informed decision-making in software project management.