Multifaceted DDoS Attack Prediction by Multivariate Time Series and Ordinal Patterns

Distributed Denial of Service (DDoS) attacks are recurrent threats, reaching unprecedented malicious network traffic volume and speed against targets. Predicting attacks is paramount to reduce costs in mitigating or remediating them. But, it is a challenging task due to attack multifaceted properties (e.g., different attack vectors and network traffic). The multidimensional nature of these attacks requires equally multifaceted defenses. Existing solutions to DDoS attack prediction employ homogeneous data sources for model training, limiting the perspective in the face of data variability. The proposal benefits from multivariate time series correlation and noise tolerance from ordinal patterns transformation, a method suitable for IoT environments. The method predicts a DDoS attack up to 35 minutes before it effectively begins, surpassing the literature relying on approaches that require labeled data and solutions based on complex neural networks.