A Review Of Training Data Selection In Software Defect Prediction
The publicly available dataset poses a challenge in selecting the suitable data to train a defect prediction model to predict defect on other projects. Using a cross-project training dataset without a careful selection will degrade the defect prediction performance. Consequently, training data selec...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Little Lion Scientific Islamabad Pakistan
2020
|
| Online Access: | http://eprints.utem.edu.my/id/eprint/24911/2/FULL%20PAPER_%20A%20REVIEW%20OF%20TRAINING%20DATA%20SELECTION%20IN%20SOFTWARE.PDF http://eprints.utem.edu.my/id/eprint/24911/ http://www.jatit.org/volumes/Vol98No12/9Vol98No12.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The publicly available dataset poses a challenge in selecting the suitable data to train a defect prediction model to predict defect on other projects. Using a cross-project training dataset without a careful selection will degrade the defect prediction performance. Consequently, training data selection is an essential step to develop a defect prediction model. This paper aims to synthesize the state-of-the-art for training data selection methods published from 2009 to 2019. The existing approaches addressing the training data
selection issue fall into three groups, which are nearest neighbour, cluster-based, and evolutionary method. According to the results in the literature, the cluster-based method tends to outperform the nearest neighbour method. On the other hand, the research on evolutionary techniques gives promising results but is still scarce. Therefore, the review concludes that there is still some open area for further investigation in training data selection. We also present research direction within this area |
|---|