An empirical evaluation of outlier deletion methods for analogy-based cost estimation
Proceedings of the 7th International Conference on Predictive Models in Software Engineering
Systematic literature review of machine learning based software development effort estimation models
Information and Software Technology
| Hi-index | 0.00 |
Accurate software effort estimation is essential for successful project management. To improve the accuracy, a number of estimation techniques have been developed. Among those, Analogy-Based Estimation (ABE) has become one of the mainstreams of effort estimation. In general, ABE infers the effort to accomplish a new project from the efforts of the historical projects which possess similar characteristics. ABE is simple, yet it can be affected by the noise in historical projects. Noise is generally the data corruptions which may cause negative affect on the performance of a model built on the historical data. In this study, we propose an approach to filtering noise in the historical projects to improve the accuracy of ABE. We introduce and measure the Effort-Inconsistency Degree (EID), the degree that the effort of a historical project is inconsistent from those of its similar projects. Based on EID, we identify and filter the noise in terms of the inconsistent historical project data. We have validated the performance of ABE with our approach and three representative filtering techniques, namely the Edited Nearest Neighbor algorithm, the Univariate Outlier Elimination, and the Genetic Algorithm based project selection, on three software project datasets (Desharnais, Maxwell, and ISBSG (International Software Benchmarking Standards Group) Telecom). The experimental results suggest that our approach can improve the accuracy of ABE more effectively than can the other approaches.