A feature matching and transfer approach for cross-company defect prediction

• Published in: The Journal of systems and software Vol. 132; pp. 366 - 378
• Main Authors: Yu, Qiao, Jiang, Shujuan, Zhang, Yanmei
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.10.2017
• Subjects: Feature matching; Feature transfer; Heterogeneous features; Software defect prediction; Feature transfer; Feature matching; Software defect prediction; Heterogeneous features
• ISSN: 0164-1212; 1873-1228
• DOI: 10.1016/j.jss.2017.06.070

•A feature matching algorithm is designed to address the heterogeneous features.•A feature matching and transfer (FMT) approach for cross-company defect prediction.•An empirical study is conducted on 16 datasets from NASA and PROMISE.•The results show that FMT is effective for cross-company defect prediction. Software defect prediction has drawn much attention of researchers in software engineering. Traditional defect prediction methods aim to build the prediction model based on historical data. For a new project or a project with limited historical data, we cannot build a good prediction model. Therefore, researchers have proposed the cross-project defect prediction (CPDP) and cross-company defect prediction (CCDP) methods to share the historical data among different projects. However, the features of cross-company datasets are often heterogeneous, which may affect the feasibility of CCDP. To address the heterogeneous features of CCDP, this paper presents a feature matching and transfer (FMT) approach. First, we conduct feature selection for the source project and get the distribution curves of selected features. Similarly, we also get the distribution curves of all features in the target project. Second, according to the ‘distance’ of different distribution curves, we design a feature matching algorithm to convert the heterogeneous features into the matched features. Finally, we can achieve feature transfer from the source project to the target project. All experiments are conducted on 16 datasets from NASA and PROMISE, and the results show that FMT is effective for CCDP.