Abstract : Distributed Denial of Service (DDoS) attacks are potential threats to network stability; network traffic for machine learning models is challenging to analyze due to its complexity. In this study, a novel hybrid complex multilayer perceptron (HCMLP) model is introduced for detecting DDoS attacks. To improve HCMLP’s feature extraction and classification capability, we adopted multibranch structures, residual connections, dense blocks, and attention mechanisms. There are three parts to the architecture: a standard MLP was used to learn the features at first, a DenseNet-type structure was used to reuse features, and ResNet-type residual blocks were used to solve the vanilla gradient problem. This approach to fusion averages outputs from all branches to create a rich set of features. Furthermore, we have improved the model's performance via reinforcement learning with specific loss functions. The CIC-IDS 2018 dataset has been used to evaluate the performance of the HCMLP. Results showed precision of 1.00, recall and F1 scores of 1.00, and accuracy of 99.96 percent. Ranking the attacks within 30 seconds, the proposed method is highly efficient and can achieve a 15.2% increase in runtime for the CIC-IDS 2018 benchmark. The UNSW-NB15 dataset has been used to evaluate the performance of the HCMLP. Results showed precision of 0.965, recall and F1 scores of 0.9645, and accuracy of 96.02 percent. We ranked the attacks within a timeframe of 21.6 seconds. In this work, we address class imbalances and shifts in attack patterns under complex network environments to advance state-of-the-art DDoS detection models.

Keyword : 3D masks, Classification Error Rate, Graph Neural Networks, Reinforcement Learning, Transformer-based feature extraction.