Improving the Accuracy of Analogy-Based Effort Estimation by Local Optimization of Attribute Weight
Hyun-Sik Cho, Yeong‐Seok Seo, Doo‐Hwan Bae · Jeongbo gwahaghoe nonmunji. so'peuteuweeo mich eung'yong · 2011
Accurate software effort estimation is essential for efficient software project management. Among the existing methods for software effort estimation, Analogy-Based effort Estimation (ABE) is one of the most commonly used methods in practice, which is a method to estimate the effort for a new project by using the effort of similar projects in historical projects. In order to improve the accuracy of effort estimation of ABE, many studies suggested attribute weighting techniques that assign weight for each attribute to identify more similar projects with a new project. However, since the existing techniques generate a global weight value that is optimized on the whole historical projects, new projects that have different characteristics use the weight value that is not optimized for each project. It may degrade the accuracy of effort estimation of ABE due to neglecting of different characteristic of each project. Thus, in this paper, we propose a locally optimized attribute weighting technique that generates attribute weight suitable and optimized for a new project to be estimated. The proposed technique is validated with three different attribute weight techniques based on the industrial data sets. The experimental results show that the proposed technique outperforms other techniques in accuracy of effort estimation of ABE.