Discovering Signals from Web Sources to Predict Cyber Attacks.

摘要: Cyber attacks are growing in frequency and severity. Over the past year alone we have witnessed massive data breaches that stole personal information of millions people wide-scale ransomware paralyzed critical infrastructure several countries. Combating rising cyber threat calls for a multi-pronged strategy, which includes predicting when these will occur. The intuition driving our approach is this: during planning preparation stages, hackers leave digital traces their activities on both surface web dark form discussions platforms like hacker forums, social media, blogs like. These provide predictive signals allow anticipating attacks. In this paper, describe machine learning techniques based deep neural networks autoregressive time series models leverage external from publicly available Web sources to forecast Performance framework across ground truth over real-world forecasting tasks shows methods yield significant lift or increase F1 top predicted Our results suggest that, deployed, system be able an effective line defense against various types targeted