Optimization of feature weights and number of neighbors for Analogy based cost Estimation in software project management

Software cost estimation affects almost all activities of software project development such as: bidding, planning, and budgeting, thus it is very crucial to the success of software project management. In past decades, many methods have been proposed for cost estimation. Analogy based cost estimation (ABE) is among the most popular techniques due to its conceptual simplicity and empirical competitiveness. In order to improve ABE model, many previous studies have focused on optimizing the feature weights in the similarity function. However, according to some prior studies, the K parameter for the K-nearest neighbor is also essential to the performance of ABE. Nevertheless, few studies attempt to optimize the K number of neighbors and most of them are based on the trial-error scheme. In this study, we propose the genetic algorithm to simultaneously optimize the K parameter and the feature weights for ABE (OKFWSABE). The proposed OKFWABE method is validated on three real-world software engineering data sets. The experiment results show that our methods could significantly improve the prediction accuracy of conventional ABE and has the potential to become an effective method for software cost estimation.