Deep Learning Models for Cyber Security in IoT Networks

Introduction

The newest developing promising technology known as the Internet of Things links everything on Earth via the internet. IoT (the technology) has the potential to enhance and assist in both our personal and professional lives, as well as our societal well-being. The internet of things, otherwise known as IoT, includes millions of smart devices all over the globe, all linked via the internet without human intervention. Unfortunately, because of this, it is vulnerable to cyber assaults just like any other network. A powerful tool for detecting cyber-attacks is an intrusion detection system (IDS) (Roopak et al., 2019). Machine learning has formed the core of all current IDS (including the state-of-the-art models). Fog computing extends cloud computing, where fog nodes are placed closer to IoT network devices to increase scalability, lower bandwidth usage, lower latency in QoS and overall enhance overall functionality. In fog-to-node computing, real IoT networks may achieve success. Figure 1 shows the architecture of a fog-to-node model, where fog nodes are dispersed and each node hosts distributed parallel computing that provides distributed fog systems with processing, control, and storing of IDS components that are more central to IoT network objects. In contrast to cloud services, fog nodes can identify a cyber-attack effectively and promptly. Internet of Things (IoT) refers to a collection of diverse technologies ranging from supercomputers to small devices with extremely little computational capacity, therefore protecting an IoT network poses major security challenges (Yuan et al., 2017). DDoS was a significant cyber-attack on the Internet of Things (IoT) that caused considerable damages. Attacks against web servers often use many hosts to overwhelming target systems, resulting in a total crash. This causes genuine users to be unable to use the system. It is expected that the denial-of-service attack would grow to around 17 million by 2020 as predicts. DL is a wider subfield of machine learning known as machine learning and machine learning (ML) is deeper than that, it is in fact a much larger network known as a deep neural network. This research, published in 2016, shows that deep learning methods such as deep belief networks and also convolutional neural networks (CNN) were developed in the 1980s and have since proved successful in a variety of areas such as image processing, natural language processing, and self-driving cars. One of the main challenges of deep learning techniques is the bigger the training dataset, the longer the training period.

Figure 1.

Depicts the architecture of an Internet of Things network from fog to node.

Figure. 2 shows the fundamental architecture of a deep learning model, which consists of one input layer followed by several hidden layers. The last layer of the model has one input and produces an output. CNN is a deep learning model that has been widely used in image and language processing (Sharif Razavian et al., 2014). The raw picture is immediately supplied to CNN models without any pre-processing, and the model then does various convolution operations on the image data (Hao et al., 2018). In the areas of NLP (natural language processing) and text processing, RNNs (Recurrent neural networks) have shown positive results. LSTM (Long Short-Term Memory) is a kind of RNN. Because LSTM is applied directly to raw data, it has a number of advantages.

Figure 2.

The Deep Learning Model's Architectural Design