An over-sampling method for analogy-based software effort estimation.
Abstract
This paper proposes a novel method to generate synthetic project cases and add them to a fit dataset for the purpose of improving the performance of analogy-based software effort estimation. The proposed method extends conventional over-sampling method, which is a preprocessing procedure for n-group classification problems, which makes it suitable for any imbalanced dataset to be used in analogy-based system. We experimentally evaluated the effect of the over-sampling method to improve the performance of the analogy-based software effort estimation by using the Desharnais dataset. Results show significant improvement to the estimation accuracy by using our approach. In this paper, we propose a novel approach to generate synthetic project cases and add them to a fit dataset. The approach described in this paper is called over-sampling method that adds synthesized projects similar to triangular plots shown in Figure 1(b). As shown in Figure 1(b), the synthetic projects like triangular plots are added in the middle of functional size to large of functional size, and then the number of projects that are similar features with target project would be increased. Hence, the oversampling could have potential for the performance improvement of the analogy-based method. However, to our knowledge, no study has reported the effects of applying the over-sampling to general effort estimation. Mittas et al. has proposed "resampling" such as iterated bagging method [6] that selects some projects from all projects and uses them as a new dataset. On the other hand, "over-sampling" means the addition of a new synthetic project based on characteristics of projects in a dataset. We have experimentally evaluated the effects of applying the oversampling method using the Desharnais dataset from a Canadian Software house. Categories and Subject Descriptors D.2.9 [Software Engineering]: Management - Cost estimation General Terms Management, Measurement, Experimentation Keywords Software effort estimation, analogy, over-sampling, empirical study 1. INTRODUCTION Section 2 introduces the over-sampling method. Section 3 describes the details of the evaluation setting. Section 4 provi