With billions of gadgets capable of sending important data about the physical world and carrying out straightforward activities, the Internet of Things (IoT) has materialised and given rise to the paradigm of always-on connectivity for anything. By 2030, billions of IoT devices will be employed across a variety of industries, including manufacturing, healthcare, transportation, and smart cities [2]. IoT device security is a major concern, yet it can be challenging to ensure that they work as intended. The heterogeneous devices connect and interact with one another as part of the Social Internet of Things (SIoT), a specialised subset of the Internet of Things, as well as socialise and work together to complete one or more specific tasks.
Social IoT, which is the core of this type of social interaction, depends on IoT devices working together. Figure 1 gives the generic process of social IoT concepts. It cannot, however, be fully realised without integrated support in the operating systems (OSs), firmware, or other software used by these IoT devices [3]. Everyday objects can connect and exchange data with other devices and computer systems via IoT networks by means of sensors, processors, applications, and other technologies that are integrated into them.These social connections enable interaction between devices and people, aid their social networks, and make information sharing easier. IoT reuses the principles and concepts of human social networking to address IoT-related issues.The models now used in human social networks can therefore be used to solve IoT-related problems [4].Although security solutions have advanced, they still have issues with scalability, centralization, and unclear design. Additionally, they overemphasise the physical elements of the gadgets while ignoring the social intelligence that comes from interactions between the gadgets and their users. Since trust is built through connections in contexts like families, enterprises, and friendships, for example, these solutions do not enhance actual human interactions. Devices connected to the Internet of Things are particularly vulnerable to network attacks such DDoS attacks, fraud, and spoofing.
Mirai is a rare type of botnet that abuses IoT devices to launch massive distributed denial-of-service (DDoS) attacks [5]. Hajime poses a serious risk to reputation and customer trust [6]. DDoS attacks are among the destructive actions carried out by these attacks, which compromise IoT devices. In this study, attack vectors and hazards like novel attacks, botnets, and other forms of cyberattacks are used.Due to customer scepticism and a hampered adoption of the IoT, manufacturers and organisations linked to compromised devices may suffer reputational damage. IoT-Bot assaults also have monetary and legal repercussions, including incident response expenses and potential legal actions. Therefore, maintaining reputation and customer confidence is essential for the long-term development and acceptance of the IoT.Denning [7] suggested creating an IDS that would use AI approaches to find anomalous flows and probable intrusions. This strategy led to the development of a new IDS branch using learning algorithms. In the past 30 years, numerous studies have investigated the use of IDSs to automatically distinguish between normal and aberrant flows in systems and networks using machine learning techniques. IDS solutions are however still impacted by high false alarm and low detection rates.
As a result of its advancements in computer vision, image processing, and natural language processing, deep learning (DL) has assisted IDSs in cybersecurity [8–10]. Several deep neural networks (DNN) have been created for IoT network intrusion detection, including convolutional neural networks [11], generative adversarial networks (GAN) [12], encoders [13], deep belief networks [14] and recurrent neural networks [15]. CNN is recognised for its incredibly high classification accuracy and capacity to self-extract essential features for a variety of variables, making it one of the most promising deep learning (DL) models. Due to the wide range of datasets, DL models not only require more memory and processing power than existing machine learning (ML) models but also have difficulty enhancing detection performance.
CNN is recognised for its incredibly high classification accuracy and capacity to self-extract essential features for a variety of variables, making it one of the most promising deep learning (DL) models. Due to the wide range of datasets, DL models not only require more memory and processing power than existing machine learning (ML) models but also have difficulty enhancing detection performance. The majority of studies therefore focused on reducing the overall complexity of DL models while enhancing NIDS detection accuracy. Over the past few years, a number of datasets have been created for intrusion detection [16–19]. The CICIDS2018 dataset, an update to CICIDS2017, is used in this study because it is extensively used [20]. The fact that it also incorporates various kinds of network traffic from real world environments is one of the key reasons for its popularity.
The major contribution of this work is to propose lightweight CNN-based network intrusion detection model with better performance than existing models for multi-class classification on various attacks from CICIDS2018 dataset. The dataset is pre-processed based on data transformation and numerical standardizationbefore training into proposed CNN model.
The article is divided into following sections. Section II gives contemporary works related to intrusion detection and prediction using various DL and ML models. Section III discusses the datasets and architecture proposed CNN model and section IV discusses the results and comparison with concluding remarks in section V.