An Intrusion Detection System Using Extended Kalman Filter and Neural Networks for IoT Networks

The unparalleled growth of the Internet of Things (IoT) is introducing a new paradigm shift in networking technology. By connecting everyday devices to the internet and enabling them to gather and share information, IoT is paving the way for embedding intelligence into them. Though it is benefiting various business domains, with numerous uses across medical, financial, and industrial sectors, the existing underlying issues of traditional networks can be observed in IoT networks. Given their diverse nature, the concerns are considerably greater, security and privacy issues being the prime concern of the research fraternity. For instance, some devices collect and transmit sensitive personal data, making them potentially vulnerable to hackers. Thereby, the security of IoT devices and networks has become an increasingly critical necessity. Recognizing that IoT attacks cannot be averted, timely detection becomes crucial. Recent research shows that machine learning techniques have proven their credibility in establishing advanced Intrusion Detection Systems (IDSs) for traditional networks. Taking this forward, it is proposed to build an IDS for IoT networks using neural networks with Extended Kalman Filter as a backpropagation strategy. Two neural network models are explored: Artificial Neural Networks and Recurrent Neural Networks. Two different datasets are used to analyze the proposed systems, namely, the NSL-KDD dataset and the BoT-IoT dataset. The proposed models are assessed and compared using metrics such as accuracy, detection rate, false positive rates, and false negative rates.