研究目的
To propose and evaluate micro interaction metrics (MIMs) that leverage developer interaction information for improving software defect prediction accuracy.
研究成果
The study demonstrates that MIMs significantly improve defect prediction accuracy when combined with existing metrics, offering a cost-effective approach to code inspection and providing developers with intuitive feedback on their behaviors. MIMs show promise for future applications in IDE-centric tools for detecting and warning of behavioral anomalies.
研究不足
The study is limited by the availability and completeness of Mylyn data, as developers may not always submit their task context or may submit only a portion of it. The representativeness of the systems and developers referenced may also be a limitation.
1：Experimental Design and Method Selection
The study employs Mylyn, an Eclipse plug-in, to capture and store developer interactions such as file editing and browsing events in task sessions. The methodology involves comparing the defect prediction performance of MIMs with existing source code metrics (CMs) and change history metrics (HMs).
2：Sample Selection and Data Sources
The study uses Mylyn task session logs from Eclipse Bugzilla between December 2005 and June 2010, involving 5,973 task sessions. The instances used are files that existed both before and after the release time P of Eclipse 3.5 (Galileo).
3：List of Experimental Equipment and Materials
Mylyn Monitor for collecting developer activities in Eclipse, Understand tool for extracting source code metrics, and Eclipse CVS repository for change history metrics.
4：Experimental Procedures and Operational Workflow
The process involves collecting Mylyn task session logs, counting post-defects, extracting MIMs, CMs, and HMs, building prediction models using machine learning algorithms, and evaluating the models' performance.
5：Data Analysis Methods
The study uses the correlation-based feature subset (CFS) for feature selection, random forest algorithm for model construction, and F-measure for evaluating prediction accuracy. The Wilcoxon rank-sum test is used for hypothesis testing.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
