Detection and Mitigation of Attacks at the Edge of IoT Networks Using Deep Learning and P4

Context: The concern for security in IoT networks is becoming increasingly common, considering the accelerated use of these devices in recent years. Problem: Traditional methods for detecting attacks in IoT networks are ineffective due to device heterogeneity. While learning approaches show promise in threat detection, the computational limitations of IoT devices make it challenging to implement these solutions. Solution: This work proposes a deep learning solution for IoT attack detection, focusing on efficient threat mitigation using the Programming Protocol-independent Packet Processors (P4) language, adaptable feature extraction for algorithm inference, and execution on SBC devices. SI theory: This work is based on General Systems Theory and Machine Learning Theory, emphasizing integrating components in IoT networks and learning for attack detection. Methods: This study uses a case study to evaluate the effectiveness of deep learning algorithms in detecting attacks in IoT networks, adopting a descriptive and quantitative approach with metrics like accuracy and F1 score. Results: In our work, the Long Short Term Memory (LSTM), Convolutional Neural Network (CNN), and Gated Recurrent Unit (GRU) algorithms achieved accuracy rates of 96.9%, 95.6%, and 93.9%, respectively. However, the CNN model was selected for the final implementation due to its superior inference time and resource consumption performance. Contributions and impact in the IS area: This work contributes to Information Systems by integrating deep learning and P4 programming for attack detection and mitigation in IoT networks, enhancing security, and promoting further research in the field.