Transfer Learning of Enhanced Convolutional Neural Network with Long Short-Term Memory for DDoS Attack Detection with Fine-Tuning Process

Recently, techniques based on deep learning have offered superior outcomes in Distributed Denial-of-Service (DDoS) attacks because of their capability to examine the complicated patterns of the traffic information in the network. Yet, the deep learning model’s efficiency in safeguarding the system from various threats depends upon the network architecture and the quality of the information. Therefore, an advanced framework is developed to overcome the shortcomings in the classical networks and to enhance the attack detection accuracy. This developed framework utilized the transfer learning procedures to detect DDoS attacks. At first, important DDoS attack data is gathered from the standard resource and offered to the detection framework. Here, DDoS attack detection is executed using transfer learning of an enhanced convolutional neural network with long short-term memory (TF-E-CNN-LSTM). Additionally, the effectiveness of the designed TF-E-CNN-LSTM system is boosted by optimizing the hyperparameters by employing the developed enhanced random variable-based piranha foraging optimization algorithm (ERV-PFOA). Thus, the developed TF-E-CNN-LSTM-aided attack detection model provides superior DDoS attack detection outcomes than the traditional approaches. Later, diverse experiments are implemented in the introduced framework to verify the effectiveness of the existing techniques. The results show that the developed HDMcCNet system attains an accuracy of 95.72% when taking the batch size as 48, which leads to swift and better reliable detection of attacks. It is shown that the suggested model achieved better execution in DDoS attack identification than traditional methods.