A HYBRID GAN–DNN FRAMEWORK FOR INTELLIGENT ANOMALY DETECTION IN IOT NETWORKS

Abstract

The Internet of Things (IoT) growth has significantly improved in a very short period that increased the network threats and risks of anomalies pattern and other threats such as exploitation attacks, denial of service (DoS), and fuzzing, which can disrupt services and compromise data security. Conventional anomaly detection algorithms, including ML and DL models, tend to fail to detect various dynamic patterns of attacks due to skewed datasets and low generalization. To address these limitations, this paper proposes an embedded and integrated Generative Adversarial Network Deep Neural Network (GAN- DNN) model of detecting abnormalities in IoT networks. An artificial data generation in a GAN is then applied to a DNN classifier, with feature optimization so that the accuracy of detection can be increased when the data is diverse because of an artificially generated realistic anomaly data. Integrated strategy performs better than standalone DNN models as it leverages the great learning capacity of DNNs and the capability of GANs to form complex data distributions. The proposed GAN-DNN model was evaluated experimentally on benchmark datasets of IoT network traffic, and its accuracy surpassed the accuracy of the baseline DNN of 0.84 with a model accuracy of 0.91. It is interesting to note that it significantly improves the recall of exploits and fuzzers and gives perfect detection of DoS attacks with performance measures such as recall, precision, and F1-score1. The effectiveness of the results shows significant improvements in terms of real-time detection of IoT network anomalies.