Optimal feature selection with CNN-feature learning for DDoS attack detection using meta-heuristic-based LSTM

Nowadays, the digital era is reshaped by new technologies, and the cyber-attacks are more sophisticated and becoming as a commonplace. The distributed denial of service (DDoS) attacks are the exponentially-growing and major prevalent attack that targets the emerging and changing computational network infrastructures around the globe. It is complex to distinguish the DDoS attack traffic from the legitimate network traffic when the transit happens from the zombies or attacker to the victim. The DDoS attack is considered as a stubborn network security conflict. Yet, these algorithms need a priori knowledge regarding the classes, and it is not possible to adapt to the subsequent varying network traffic trends in an automatic manner. This creates the requirement for the enhancement of the novel DDoS detection mechanisms that in turn sophisticated and targets the DDoS attacks. The main intent of this paper is to implement the DDoS detection model through deep learning by the integration of convolutional neural network (CNN), and optimized long short-term memory (LSTM), so called CNN-O-LSTM. On the standard five benchmark datasets, the optimal feature selection is performed by the closest position-based grey wolf optimization (CP-GWO) with the consideration of minimizing the correlation among the features. With the optimally selected features, CNN is adopted for the feature learning process, from which the features of the second pooling layer are extracted, which is used for performing the detection. The adoption of optimally selected features with the CNN features enhances the detection performance with the most significant features. Finally, the optimized LSTM is used in the detection phase, which aims to maximize the detection accuracy by optimizing the hidden neurons of LSTM. The proposed DDoS detection scheme is experimented on a set of benchmark datasets, and the outcomes are compared over the traditional models.