A Deep Learning Framework by Leveraging CNN and LSTM Hybridization for Effective DDoS Attack Detection in IoT Environments
Shanmugam Sathiya Devi, K. Lalitha · 2025
Distributed Denial of Service (DDoS) attacks represent the biggest threat towards the security and stability of IoT networks, and they can even cause a server crash, severely affecting services for users. Increased growth in IoT networks has accelerated a greater demand for efficient and accurate detection mechanisms for DDoS. Machine learning and deep learning have been powerful techniques that advance the state of detecting such attacks. We propose, in this paper, the novel concept of a hybrid model between CNN and LSTM to detect DDoS attacks effectively in the IoT environment. To extract spatial features, the architecture comprises of three convolutional layers with kernel sizes of 3, neuron count of 64, neuron count of 128, and neuron count of 256. These layers were followed by LSTM layers that process sequential dependencies in the data for better detection as they learn temporal patterns. The evaluation was done on the CICIOT 2023 dataset. These experiments have shown that the proposed hybrid CNN-LSTM model can be achieved to get an accuracy of 96.78% in training and 96.08% during testing, outperforming all the traditional methods considered above. The above results establish the model as effective in the detection of DDoS attacks on IoT environments, providing a robust real-time intrusion detection solution. This work proves that DNN models can provide highly accurate and robust network security from DDoS attacks while enhancing the reliability and stability of IoT networks.