A Survey of Automatic Software Vulnerability Detection, Program Repair, and Defect Prediction Techniques

Table 6

Technical characteristics of within-defect prediction methods.

System/writer	Datasets	Metrics	Feature generation	Data labeled
Wang et al. [43]	PROMISE	P F1 Recall	Parse source code, handle noise, and map tokens, generate feature via DBN	CLNI
Dam et al. [44]	Samsung	P F1 Recall	Parse source code, map AST nodes, generate feature via Tree-LSTM	Model generation
DP-CNN [45]	PROMISE	P F1 Recall	Parse source code, extract and encode token, generate feature via CNN	Repository provided
SDNN [46]	NASA	F1 AUC	Delete repeated entities, replace missing value, data normalization	Repository provided
CAP-CNN [47]	PROMISE	F1	Split source modules, encoded as vector via pretrained word2vec, generate feature via CNN	Repository provided
DefectLearner [48]	12 open source projects	P F1 Recall	Remove comment, use word embedding method, generate feature via LSTM	Projects provided