Comparative analysis of soft computing techniques for predicting software effort based use case points. Issue 1 (1st February 2018)

• Record Type: Journal Article
• Authors: Azzeh, Mohammad; Nassif, Ali Bou; Banitaan, Shadi
• Is Part Of: IET software. Volume 12:Issue 1(2018)
• Page Start: 19
• Page End: 29
• Publication Date: 2018-02-01
• Subjects: software metrics -- neural nets -- fuzzy reasoning -- support vector machines -- regression analysis -- software management -- project management -- systems analysis; soft computing -- software effort prediction -- use case points -- UCP -- software project -- requirements phase -- analysis phase -- software size metrics -- use case driven development -- software industry -- adaptive neuro fuzzy inference system -- support vector regression -- regression neural network; Computer software -- Periodicals; Software engineering -- Periodicals; 005.1
• DOI: 10.1049/iet-sen.2016.0322
• Languages: English
• ISSNs: 1751-8806

Abstract

Abstract : The size of a software project is a key measure of predicting software effort at the requirements and analysis phase. Use case points (UCP) is among software size metrics that achieved good reputation because of the increasing popularity of use case driven development methodologies in software industry. Nevertheless, there is no consistent method that can effectively translate the UCP into its corresponding effort. Previous estimation models were built using a very limited number of projects, and they were not well examined. The soft computing techniques were rarely applied for such problem and their performances have not been well investigated using a systematic procedure. This study looks into the accuracy and stability of some soft computing methods for the problem of effort estimation based on UCP. Four neural network methods, adaptive neuro fuzzy inference system and support vector regression have been used in this comparative study. The results suggest that most used soft computing techniques can work well with good accuracy for such problem. Among them, the general regression neural network is the superior one with stable ranking across different accuracy measures. Also, it has been found that using adjustment variables with basic UCP variables, solely or together, have positive impact on the accuracy and stability.