Discriminating Flash Events From DDoS Attacks Using Cross-Channel Attentional Deep Network

Distributed denials of service (DDoS) attacks are the core attacks in the cyber world, where the attackers intentionally transmit more attack packets to the normal traffic to disrupt system services. Flash events on the other hand represent an unexpected surge, in which the traffic is quite different from illegitimate traffic. Discriminating flash events from DDoS is an important task to identify legitimate traffic from illegitimate traffic, but remains challenging due to certain difficulties in detection and generates maximum false alerts in flash events. To address these issues, this research proposed a deep recurrent neural network with a hybrid efficient cross-channel attention module (EC-deep RNN) to effectively discriminate flash events from DDoS. This model is ensemble with a hybrid efficient cross-channel attention module, which is the combination of two channel-based attention techniques, namely efficient channel attention as well as cross-channel attention that boost the model’s progress for better performance. The EC-deep RNN model provides significant results and also eliminates existing limitations including over-fitting issues, poor solutions, and complications of using large datasets. Moreover, the model achieves 95.50% accuracy, 95.94% precision, 95.83% of recall, and 95.89% of F1-score compared to other methods.