Threshold Based Neighborhood Selection for Case-Based Reasoning in Software Effort Estimation
Qin Liu, Jiakai Xiao, Hongming Zhu · 2017 International Conference on Computer Systems, Electronics and Control (ICCSEC) · 2017
Purpose: Case Based Reasoning (CBR) a widely studied method for software effort estimation (SEE). In this paper, we propose the Threshold based Neighborhood Selection (TNS) method to address the issue of noise and outliers in datasets. Method: Specifically, TNS first learns project similarity threshold from historical projects. Then for a project, neighbors whose similarity with the current project falls within learned threshold are selected for estimation. Results: Experiments on 6 benchmark datasets demonstrates that TNS achieves optimal MAR (Mean Absolute Residual) on 4 out of 6 datasets and second and third optimal on the other 2 datasets. Conclusion: The results prove that TNS is an appropriate method to address the issue of noise and outliers in datasets.