Internet of Things Intrusion Detection: A Deep Learning Approach

Abstract

The Internet of Things (IoT) is a shifting paradigm that allows the integration of billions of devices with the Internet. With its wide range of application domains, including smart cities, smart homes, and e-health, the IoT has created new challenges, particularly security threats. Traditional security solutions, such as firewalls and intrusion detection systems, need amending to fit the new networking paradigm. Given the recent advances in machine learning, we investigated the use of deep learning algorithms for anomaly detection. The IoT collects a massive amount of data from the environment, and deep learning is based on a set of algorithms striving for the data. Intrusion detection systems are used to expose network threats and are an effective means of protecting network assets. Anomaly detection is a conventional intrusion detection approach that separates normal and abnormal network traffic using statistical, rule-based, or machine learning models. Of the machine learning models, deep learning is a neural network algorithm that has provided breakthroughs in domains such as object and voice recognition. However, there are limitations in applying deep learning to network anomaly detection. This paper proposes a novel anomaly detection framework based on unsupervised deep learning algorithms for revealing network threats. Our research explores the applicability of deep learning to detect anomalies by evaluating the use of Restricted Boltzmann machines as generative energy-based models against Autoencoders as non-probabilistic algorithms. The study provides an in-depth analysis of unsupervised deep learning algorithms. The simulations studies show $\approx$ 99% detection accuracy, which is significantly improved compared to the closely related work.