An Optimization-Based Method to Increase the Accuracy of Software Development Effort Estimation
Vahid Khatibi Bardsiri, Amid Khatibi, Elham Khatibi · 2013
Software development effort estimation has become a challenging issue for developers, managers and customers during the last years. Some of the reasons behind this challenge are inconsistency of software projects, complexity of production process, intensive role of humans, unclear requirements and so on. In order to remedy the challenge, quite many estimation methods have been proposed in the last decades; and the attempts to increase the accuracy of estimates have not been stopped yet. Among all the existing estimation methods, analogy based estimation (ABE) is the most popular one that has been extensively used in field of software development effort estimation. This is because ABE is applicable to be used at early stages of software projects based on a smooth and clear estimation process. Despite advantages, ABE is confronted by high number of outliers and irrelevant projects frequently appeared in software project datasets. This paper proposes a hybrid model in which the genetic algorithm and ABE are combined to make a high performance estimation model. Indeed, the process of attribute weighting is adjusted so that the performance of ABE is improved. A real dataset is utilized to evaluate the accuracy of the proposed hybrid model. The comparison between the proposed model and different types of ABE certified that the performance metrics have been improved by the proposed model.