This section looks at some recent research findings pertinent to IoT network security and intrusion detection approaches. Based on the performance outcomes, it also examines the benefits and drawbacks of the current systems.

Da Costa et al. [18] presented a comprehensive review examining several machine learning techniques for detecting intrusions from IoT networks. This review’s focus is on various machine learning and evolutionary computation-based processes. This article aims to enlighten readers on the state of the field’s literature and serve as a new resource for academics looking into IoT security challenges. The work includes several innovative approaches to intrusion detection and network security in digital environments. Although these methods aim to increase intrusion detection recognition rates, it is believed that the false positive rate continues to pose an issue that must be dealt with across all research studies.

Islam et al. [19] intended to spot the different types of IoT threats by using a set of machine learning and deep learning algorithms, including support vector machine (SVM), decision tree (DT), random forest (RF), long short-term memory (LSTM), and deep belief network (DBN). Data analysis-based strategies are used in this study since they are quicker to implement than alternative methods and more effective for addressing events that are not yet known to have resulted from known attacks. The framework’s primary goal is to create an intelligent, safe, and reliable IoT system that can identify its vulnerabilities, act as a safeguarding barrier for severe cyberattacks, and recover itself. This article suggests a learning-based methodology that can detect and guarantee the system’s security under unusual circumstances. Nimbalkar et al. [20] investigated the different feature selection techniques for developing an effective intrusion detection system (IDS) to protect IoT networks. Using an information gain (IG) and gain ratio (GR) selection models reduces the total count of features to 50%, which is provided to the JRip classifier for accurate detection. Moreover, the authors aim to generate a compact dataset after completing the data preprocessing operations. In addition, the accuracy of the suggested classifier is validated with and without IG and GR feature selection techniques. This study’s results indicate that the classifier’s accuracy is significantly increased with these feature reduction techniques.

Hindy et al. [21] conducted a detailed case study to investigate the different machine learning techniques used for intrusion detection and classification. In this work, the MQTT-IoT-IDS 2020 dataset has been used as the primary source for processing, where the uni-directional and bi-directional features are considered for an efficient classification. Alsaedi et al. [22] utilized the data-driven approaches for detecting intrusions from IoT and IIoT networks using the telemetry ToN-IoT dataset. The authors aim to develop a new dataset for protecting large-scale networks from recent cyber threats. Moreover, it investigated some of the famous and emerging datasets used in this field for ensuring security, which are listed in Table 1.

Zhou et al. [23] deployed a graph neural network (GNN)-based intrusion detection methodology for IoT systems. Here, the hierarchical adversarial attack (HAA) generation algorithm has been developed to detect unknown attacks more accurately. In addition, the random walk with restart (RWR) technique is also deployed to identify the vulnerable nodes from the network by computing the priority level. This study uses the open-source UNSW-SOSR 2019 dataset to investigate the suggested model by comparing it with some of the baseline techniques. However, this mechanism consumes more time to predict the solution, which could be the major drawback of this work. Wahab et al. [24] introduced an online deep learning approach for spotting intrusion from IoT networks. In this study, the data drift detection technique has been used to identify the intrusion data streams by analyzing the variance of features. The suggested method ensures sustainable performance with effective intrusion detection solutions. The advantage of this mechanism was that it supports both drift detection and adoption for intrusion identification and recognition. Yet, the suggested deep neural network technique has high computational and time complexity.

Kan et al. [25] utilized an adaptive particle swarm optimization (PSO) technique incorporated with CNN algorithm for identifying network intrusions from IoT systems. Also, it focuses on the development of a multi-type of intrusion detection model with improved prediction probability and reliability. Moreover, the highest probability prediction is carried out using cross-entropy loss function. The suggested APSO-CNN intrusion detection results are estimated according to the training loss and accuracy measures.

Abdalgawad et al. [26] implemented a bi-directional generative adversarial network (Bi-GAN) model for spotting cyberattacks from the IoT 23 dataset. In this study, the adversarial autoencoder has been used along with the Bi-GAN model for spotting a variety of attacks from the network. In addition, a 10-fold cross-validation strategy is used to test and validate the performance and outcomes of the suggested mechanism. According to the study results, it is noted that the generative deep learning techniques provide high accuracy with better detection performance.

Kumar et al. [27] developed a distributed intrusion detection system for spotting DDoS attacks from IoT networks. Typically, a security mechanism must handle large amounts of data generated by IoT devices in a distributed way and apply the proper statistical methods in a viable architecture. Shukla et al. [28] applied an artificial intelligence (AI)-based intrusion detection methodology for IoT security. The work examined the usage of several machine learning methods inside IoT. It also highlighted how crucial it is to select the right data for a model [29]. There is also a discussion of the naive Bayesian network intrusion detection technique. Most notably, it looked at how crucial it is to prepare IoT system events for classification, and it examined the consequences of utilizing a hidden naive Bayes multi-classifier, discovering that it outperformed the standard model. According to the AI paradigm [30], every single node in the IoT network has an internal AI process that is only in charge of that node. This method can be scaled because the total number of sessions grows according to the number of analyzed nodes. In [31], the authors introduce a hierarchical fog computing-based architecture for Industry 5.0’s smart energy-supplying systems. It efficiently handles data-intensive analysis from IIoT devices, outperforming traditional cloud computing, and ensures data security through attribute-based encryption (ABE).

Albasheer et al. [32] looked at network intrusion detection systems (NIDS) and the problems they have, such as a lot of false positives, different ways to deal with alert correlation, and insights into network security. Researchers in [33] introduced an AI-driven cross-platform VPN system for detecting and categorizing attack risks. They demonstrated the efficiency of an extended gradient boosting (XgBoost)-based AI algorithm in preventing cyberattacks and integrating with a Cassandra big data system. In [34], researchers proposed an on-the-move intrusion detection system (OMIDS) for electric vehicle networks, showing high classification accuracy for various threats. In [35], the authors explored the potential of supervised machine learning for IDS, achieving high classification performance. In [36], the authors introduced “IntruDTree”, a machine learning-based intrusion detection system designed for the IoT.

According to Table 2, it shows that some strategies can lower the false positive rate, but in doing so, they also require more training and labeling. However, some methods reverse the process, reducing the false positive rate at the expense of high computational costs for both training and testing. Such a problem is highly pertinent to intrusion detection because real-time detection is an important consideration.