N THE PREDICTABILITY OF SOFTWARE EFFORTS USING MACHINE LEARNING TECHNIQUES

摘要：

This paper investigates the predictability of software effort using machine learning techniques. We employed unsupervised learning as k-medoids clustering with different similarity measures to extract natural clusters of projects from software effort data set, and supervised learning as J48 decision tree, back propagation neural network (BPNN) and na¨ive Bayes to classify the software projects. We also investigate the impact of imputin missing values of projects on the performances of both unsupervised and supervised learning techniques. Experiments on ISBSG and CSBSG data sets demonstrate that unsupervised learning as k-medoids clustering has produced a poor performance in software effort prediction and Kulzinsky coefficient has the best performance in software effort prediction in measuring the similarities of projects. Supervised learning techniques have produced superior performances in software effort prediction. Among the three supervised learning techniques, BPNN produces the best performance. Missing data imputation has improved the performances of both unsupervised and supervised learning techniques.

关键词： Software effort prediction K-medoids BPNN Data imputation

作者: Wen Zhang Ye Yang Qing Wang

作者单位: Laboratory for Internet Software Technologies, Institute of Software, Chinese Academy of Sciences, Beijing 100190, China

会议类型: 国际会议

会议名称: 13th International Conference on Enterprise Information System(第13届企业信息系统国际会议 ICEIS 2011)

会议地点: 北京

会议语种:英文

页码: 1645-1654

在线出版日期: 2011-06-08（万方平台首次上网日期，不代表论文的发表时间）