An IoT Intrusion Detection Method based on Focal Loss in Variational Auto-Enoders

Internet of Things (IoT) sensors continuously generate large volumes of heterogeneous and imbalanced traffic on their networks. In this traffic, the number of attack samples are very small and insignificant compared to normal samples, so the attack traffic may be wrongly classified as normal traffic and cause distributed denial of service (DDoS) attack in IoT networks. In this research, an intrusion detection system (IDS) has been proposed, using a combination of a deep variational auto-encoder (VAE) algorithm and a random forest (RF) classification, which uses the focal loss function to reduce the effect of imbalance network traffic and increase the accuracy of anomaly detection. The proposed method has been applied on a traffic network data named CICIDS۲۰۱۷ which is available online. The proposed system is able to detect attacks with up to ۹۹.۸۶% accuracy when handling the imbalanced class distribution with fewer samples, making it more convenient in real-time data fusion problems that target data classification.