Table 1.
Summary of JIT defect prediction works by using feature learning.
Fig 1.
Graph-based ML JIT defect prediction pipeline.
Fig 2.
Contribution graph and its projection.
a) A toy contribution graph. Check marks represent clean changes, whereas cross marks represent defect-prone changes. (b) Corresponding one-mode projection on the developer side. Classification is driven by developer-based features alone as the one-mode projection graph captures the connectivity around both endpoints (i.e., developer and file) in the contribution graph.
Table 2.
Summary of notation used.
Table 3.
The 14 open-source target projects from the dataset.
Table 4.
Classification results for both settings with the two best performing classifiers shown in bold text.