WCSE

Redundancy Weighting for Software Effort Estimation with Case Based Reasoning

Abstract— Case-based reasoning (CBR) is a widely used approach in software effort estimation (SEE). Unfortunately, it may be over fed by redundant feature(s) that may lead to erroneous prediction. To alleviate the problem, this paper proposes a Relevance-Redundancy (R2D) distance that incorporates redundancy weighting. Experiment results demonstrate that R2D achieves optimal MAR and Pred(25) on 4 benchmark datasets with an average improvement of 17.4%and 27.8% against second optimal methods.

Index Terms— feature weighting, redundancy, mutual information, case based reasoning

Qin Liu, Hongming Zhu
School of Software Engineering, Tongji University, CHINA
Jiakai Xiao
Department of Computer Science and Technology, Tongji University

[Download]

Cite: Qin Liu, Jiakai Xiao, Hongming Zhu, "Redundancy Weighting for Software Effort Estimation with Case Based Reasoning," Proceedings of 2017 the 7th International Workshop on Computer Science and Engineering, pp. 181-187, Beijing, 25-27 June, 2017.