Abstract

With increased Internet of Things (IoT) devices, network infrastructure has become very problematic with regard to security. The given paper introduces a small hybrid architecture of Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) intrusion detector in fog computing setting. Our model is assessed on the Bot-IoT data, reaching a 97.47 per cent accuracy on five types of attacks (DDoS, DoS, Reconnaissance, Theft and Normal traffic) based on 14.7 million samples. The proposed system is up to 52.54 per cent better in performance compared to standalone CNN (52.54 per cent) and is also comparable in performance with RNN-only-based architectures (97.42 per cent) and uses less energy, which is appropriate in the resource-constrained fog nodes. Meanwhile, our solution has a false positive of 2.53 percent and resolves deployment on Raspberry Pi 3 hardware, with a 6.12W maximum consumption during prediction stages. Its network traffic processing latency is below 8 seconds, which is appropriate in IoT networks to detect intrusion in real time.