Detecting vulnerability in source code using CNN and LSTM network

Automated vulnerability detection has become a research hot spot because it is beneficial for improving software quality and security. The code metric (CM) one class of important representations in source code. implicit relationships among different attributes have not been sufficiently considered traditional based on CMs. In this paper, view the local perception capability convolutional neural network (CNN) time-series prediction long short-term memory (LSTM), we propose VulExplore, compound model that consists CNN feature extraction an LSTM deep representation. Moreover, to further indicate features code, reconstruct CM dataset includes two additional attributes: maintainability index average number vulnerabilities committed per line. Our proposed numerical method can obtain both false-negative rate (FNR) false-positive (FPR) under 20% and, meanwhile, achieve recall precision over 80%, respectively.