A Scalable Meta Learning-Based Model to Secure IoT Networks

IoT plays an important role in the field of communication and networking, where there are huge amounts of useful information being transmitted. However, the high degree of diversity in IoT devices makes them untrustworthy. Machine learning techniques have resulted in noteworthy improvements in IoT security and intrusion detection. New attacks remain challenging, however, since there are few or no samples to learn the models. This is where meta-learning, a technique that uses the outputs and metadata of machine learning algorithms, may be of assistance. This article proposes a scalable model combining edge models where each can only detect a certain number of attacks and a meta-data-based learning model to detect all kinds of attacks available on all IoT edges. Our evaluation considers the model's accuracy, precision, recall, and F1 score. Both the training and prediction times and the performance of the target domains have been further improved.