Phase-wise project effort estimation using neural networks

Abstract

Inaccurate project effort estimation is a significant cause of missed project deadline, low profit, and lost bids. Many existing estimation approaches tend to predict an overall effort of a project. This study proposes a phase-wise effort estimation approach to predict an effort an effort of each phase of a project using fine-grained analysis to increase project visibility. The proposed approach is carried out in four steps. The first step concerns collection of project data from software development organizations. The second step transforms qualitative data to quantitative data. The third step finds only relevant features of a project for an effort estimation. The last step establishes a phase-wise effort estimation model using neural networks. The results show that the proposed approach utilizes only few necessary and relevant features to yield more accurate estimation than existing approaches by means of accuracy metrics and statistical hypothesis tests in performance evaluation and cross validation. Contributions of the proposed phase-wise estimation could benefit project managers in fine-grained project monitoring and control