Vulnerability Time Series Prediction Based on Multivariable LSTM

Vulnerabilities have been widely exploited to launch various cyberattacks, and become one of the most popular network security problems. Several kinds of research on vulnerabilities have been carried out, such as, mining vulnerability, tracing vulnerability, forecasting vulnerability, and so on. Vulnerability forecasting models make the prediction of zero-day vulnerabilities possible, hence they can help us learn more about the number of vulnerabilities in future days and take defence measures in advance. However, most of these models stem from statistics, which cannot adapt to our application scenario very well. Unlike traditional statistical methods, we propose a vulnerability forecasting method based on multivariable LSTM and carry on experiments on the NVD data set. In comparison with ARIMA, our method perform better in number prediction of vulnerabilities.