Comprehensive AI-Powered Strategies for Enhancing IoT Device Security: Tackling Challenges and Exploring Innovations

The expansion of the Internet beyond conventional computing devices to a wide range of everyday objects has fundamentally transformed the interaction between people and technology, with the concept of the Internet of Things (IoT). This development has led to advances such as the emergence of smart cities, industrial automation, and smart homes. However, along with these developments, significant security risks have emerged due to the diversity of operational contexts and inherent vulnerabilities of these devices. Traditional security methods are unable to solve such challenges, and the need to use innovative technologies such as artificial intelligence to enhance IoT security is clearly felt [1].

Artificial intelligence is an effective approach to countering cyber threats, such as malware, and has the ability to analyze huge amounts of data in real time to identify and mitigate security risks using Deep Learning (DL) and Machine Learning (ML) techniques [2]. The capabilities of AI have made this technology an effective solution for strengthening security in IoT devices. With the advent of the Internet of Things (IoT), the digital world has been elevated to a new level of interaction between humans and technology. From smart devices that have become an integral part of our lives, to cities and industries that promise a smart and connected future, this technology has not only provided unprecedented efficiency but also convenience. But behind these brilliant developments, there is also a shadow of widespread security risks. IoT devices, with all their advantages, may become a weakness of modern systems due to the failure to adapt traditional security methods to their specific needs [3].

With the advancement of methods such as deep learning and machine learning, AI can identify and minimize potential threats at an impressive speed. This paper attempts to explore the methods that can be used to address specific security concerns associated with IoT devices using AI. In addition to providing a comprehensive analysis of various AI-based approaches, this research also evaluates the success of these approaches. The main focus of the paper is to identify the strengths and weaknesses of AI-based security approaches and to provide suggestions for improving these approaches in the future.

Due to a number of reasons, IoT devices are characterized by problematic security:
A. Resource constraints: A great number of IoT devices suffer from constrained computing resources, e.g., memory, storage, and processing power, which make the installation of resource-intensive security measures difficult, leading to their vulnerability to cyber assaults [4]. In addition, given that a great number of IoT devices suffer from low power supply, the application of standard security measures requiring constant processing and monitoring is subjected to limitations [5].
B. Heterogeneity: One may define IoT ecosystems as a set of diverse range of devices characterized by different operating systems, communication protocols, and designs. Such variation restrains the use of standard security solutions, and maintaining the complete security of the network becomes difficult [6]. Different devices, ranging from complicated industrial machinery to basic sensors, may need customized security options. This leads to more complicated IoT security management [7].
C. Scalability: The significant scaling issues of IoT devices are due to their sheer quantity. The scalability of conventional security options may not suffice to resolve the necessities of developing IoT networks. This may potentially lead to the creation of security holes [8]. In addition, due to the dynamic nature of IoT settings, in which different devices are often withdrawn/added, adaptive security solutions are needed with the capability of growing with the network [9].
D. Privacy concerns: The fact that private personal information and business-related data are typically collected by IoT devices raises important privacy concerns. One of the key challenges is to ensure that the data are safely stored, transferred, and collected while the privacy rights of users remain secured. Given that illegal accessibility to personal information may lead to far-reaching effects, privacy is regarded as an issue with utmost importance in IoT security [10].
E. Trust: In IoT systems, trust and confidence management should be considered as critical issues in order to maintain secure and reliable interactions between people and devices. Effective trust management frameworks can cover various trust-related issues, such as data integrity, user assurance, and device reliability. Balancing privacy and trust is a challenging task because maintaining openness for accessibility while protecting user information requires complex and precise approaches, for example, anonymization [11]. Additional issues and challenges may arise in situations such as edge computing environments, where trust must be managed across remote systems [12].
F. Digital forensics: In the IoT, digital forensics is a critical issue for the analysis of security breaches by linked devices. This necessitates gathering and analyzing information from a great number of IoT devices. It may become a challenging task due to the limitations and diversity of devices [13]. Forensic data analysis is tasked with the reconstruction of the events and identification of all malicious activities by employing methods such as pattern recognition and chronology analysis [14]. In order to verify the acceptability of evidence, it is necessary to maintain legal compliance, which involves respect for privacy standards and complete recordkeeping [15].

AI-based approaches are creatively addressing many of the security concerns that IoT devices face. In this section, we briefly introduce various AI-based strategies for use in enhancing IoT security.

Machine learning for the detection of anomalies

One of the important aspects of IoT security is the anomaly detection. This is because it includes the detection of deviations from typical activities that may constitute a security issue. One may use machine learning algorithms to detect these abnormalities through learning by referring to earlier data and the recognition of the patterns that may indicate potential assaults [16].
a) Supervised learning: In supervised learning methods, labeled datasets with known outcomes are used in order to train models. By employing the attributes obtained from the data, it becomes possible for the algorithm like Support Vector Machines (SVMs) and Random Forests to discern normal network traffic from abnormal traffic [17]. In settings in which labeled attack data are limited, such techniques can be limited. This is because they demand huge amounts of labeled data for the purpose of training.
b) Unsupervised learning: Given that unsupervised learning approaches do not demand labeled data, they are helpful for the identification of unknown or new dangers. Without earlier knowledge of the attack types, techniques such as k-means clustering and Principal Component Analysis (PCA) can detect outliers and patterns in data [18]. Such methods are particularly helpful for the discovery of new dangers that otherwise could not be detected.
c) Ensemble learning: In ensemble learning, by combining a number of machine learning models, the overall performance and accuracy are improved. The methods such as bagging, stacking, and boosting can enhance the anomaly detection systems’ resilience by using the benefits of a number of algorithms [19]. Ensemble techniques are capable of increasing detection accuracy and offering a more complete perception of network dynamics.

An important aspect of IoT security is anomaly detection. This is important because it involves detecting deviations from normal activity that may indicate a security issue. Machine learning algorithms can be used to detect these anomalies by learning from previous data and recognizing patterns that may indicate potential attacks [16].
a) Supervised learning: In supervised learning methods, labeled data sets with known outcomes are used to train models. Using features obtained from the data, algorithms such as Support Vector Machines (SVMs) and random forests can be used to distinguish normal network traffic from abnormal traffic [17]. In settings where labeled attack data is scarce, such techniques can be used. This is because these algorithms require large amounts of labeled data to train.
b) Unsupervised learning: Since unsupervised learning approaches do not require labeled data, they are useful for identifying unknown or novel threats. Without prior knowledge of attack types, techniques such as k-means clustering and Principal Component Analysis (PCA) can identify outliers and patterns in the data [18]. Such methods are particularly useful for detecting new and unknown threats.
c) Ensemble learning: In ensemble learning, by combining a number of machine learning models, the overall performance and accuracy can be improved. Methods such as bagging, stacking, and boosting can increase the flexibility and resilience of anomaly detection systems by taking advantage of the advantages of a number of algorithms [19]. Ensemble techniques are able to increase the accuracy of detection and provide a more complete understanding of what is happening in the network.

Deep learning for prediction of threats

Deep learning models, particularly Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN), can effectively process complex data and predict potential threats. Such models are able to detect subtle patterns in large data sets and adjust as attack vectors change.
a) Convolutional Neural Networks (CNN): These networks are suitable for analyzing image and spatial data and are used to identify patterns and features in the input data. Convolutional networks are able to extract hierarchical features from raw data using a large number of layers of convolutional filters. This increases the accuracy in detecting complex attacks. Studies have shown that CNNs outperform standard learning techniques in a number of cybersecurity applications.
b) Recurrent Neural Networks (RNNs): These networks, especially Short-Term Memory Networks (LSTMs), are tasked with managing sequential data and are very effective for evaluating time-series data received from IoT devices. These networks are able to model temporal relationships and identify trends in network traffic and can detect persistent attacks. Long-term short-term memory networks are very useful in retaining data over long periods, making them effective for identifying complex attack patterns.
c) Generative Adversarial Networks (GANs): These networks consist of two neural networks-discriminative and generativethat compete with each other to improve performance. These networks are able to generate artificial attack scenarios and improve the training dataset, which allows for the creation of stronger security models. GANs also improve security model generalization and the training process by simulating different attack scenarios.

Reinforcement learning for adaptive security

Using Reinforcement Learning (RL), adaptive security systems can be created that evolve and learn based on their interactions with the environment. Reinforcement learning agents are able to optimize security techniques and rules through trial and error, which continuously increases the efficiency of such systems [20].
a. Model-free RL: In model-free RL approaches, for example, Deep Q-networks (DQNs) and Q-learning, optimal policies are developed without the need for an environmental model [21]. These approaches can be used in dynamic security contexts where the agent can learn through direct feedback and interaction. Model-free RL is best suited for contexts with high uncertainty and variability.
b. Model-based RL: In model-based RL, a model of the environment is learned that is used to plan and simulate actions [22]. Such a technique could be useful for IoT security, as it enables policy optimization and proactive threat mitigation based on simulated scenarios. Model-based RL increases the efficiency of the decision-making process and adapts to evolving security risks.
c. Multi-Agent RL: A great number of RL agents are included in Multi-Agent Reinforcement Learning (MARL) that work together in order to reach a shared objective. Regarding IoT security, one can use MARL to coordinate the activities of a variety of devices to jointly fight against threats. Thanks to collaborative learning, MARL enhances the IoT networks’ robustness [23].

The present section contains a variety of case studies giving practical viewpoints into the execution and effectiveness of the security measures of AI-driven IoT devices and depicting realworld difficulties, solutions, and applications for the purpose of IoT security. We hope to present the practical effect of a variety of trust management approaches, security measures, and digital forensics methodologies through the study of various tactics and scenarios. Such examples present invaluable lessons and emphasize the everchanging environment of IoT security, showing how one can use theoretical basics in real scenarios and the real-world impacts of various approaches.

Smart home security

As an important use of AI-powered security solutions, smart homes include various IoT devices, e.g., security cameras, lighting systems, and smart thermostats. AI techniques have been utilized in order to monitor and defend such network devices against various dangers.
A. Deep learning for the detection of intrusions: A study by Haddadpajouh et al [24]. reviewed the use of deep learning models for the detection of intrusions in smart home scenarios. The authors utilized an RNN so as to monitor the network data and spot malicious activities with high precision [25]. Such a strategy significantly enhances the detection of complex assaults in comparison with the currently used techniques.
B. Federated learning for privacy preservation: In order to resolve privacy concerns, smart home security employs federated learning, which enhances security and, at the same time, protects users’ privacy by training models on decentralized data collected from a variety of devices [26]. Such an approach enables collaborative learning without the need to disclose sensitive data, which makes it a suitable alternative for the purpose of privacy-sensitive situations.

Industrial IoT (IIoT)

Industrial Internet of Things (IIoT) systems, which include critical infrastructures such as energy and manufacturing systems, face specific security concerns. These concerns are compounded by the high importance of these infrastructures and the potential impact of cyber-attacks on them. To address these challenges, AIbased security frameworks have been developed to secure IIoT networks against cyber-attacks.
a) Machine learning (ML)-based anomaly detection in industrial control systems: One of the important applications of AI in IIoT is the use of machine learning-based anomaly detection to protect Industrial Control Systems (ICS). Researchers are using ML approaches to detect unusual activities in ICS networks, which helps identify potential risks and reduce downtime [27]. This technique significantly improves the reliability and security of industrial processes and enables rapid response to threats.
b) Deep learning for predictive maintenance: DL models are also widely used for predictive maintenance in IIoT contexts. These models can analyze sensor data to predict equipment failures and identify potential vulnerabilities before equipment is deployed [28]. Using this proactive strategy not only helps prevent costly downtime, but also maintains the integrity of industrial systems.

The present section explores the AI techniques geared to resolve specific IoT security challenges on the basis of the overview presented in the second chapter. By focusing on resource heterogeneity and constraints, we aim to show how advanced solutions in artificial intelligence can enhance the efficiency and security of IoT systems.

Resource Constraints

Since Internet of Things (IoT) devices typically have limited processing capabilities, it is difficult to implement comprehensive security measures on these systems. This limitation in computational resources can become a major challenge in protecting the security of devices and sensitive data. To address these issues, researchers have developed lightweight AI methods and algorithms that are specifically designed for use in resource-constrained environments.
A. Lightweight machine learning algorithms: Lightweight Machine Learning (ML) algorithms, such as linear classifiers and decision trees, are designed to provide high performance on devices with minimal resources. These algorithms are employed to balance resource utilization and optimal performance. Thus, they can enable the implementation of strong security measures without imposing additional burden on constrained systems [29].
B. Edge computing: Edge computing is also a key solution in IoT security. In this model, data is processed close to the source, i.e. on the edge devices themselves. This helps reduce the need for frequent communication with central servers. By implementing AI models on edge devices, real-time threat detection and response is possible, eliminating the need for centralized processing [30]. In addition, edge computing helps increase the scalability and efficiency of IoT security solutions. By processing data closest to the source, the time is almost dramatically reduced and threats can be responded to quickly.

Heterogeneity

Due to the diversity of IoT devices, it is difficult to use standardized security measures. The AI techniques, e.g., multi-task learning and transfer learning, present solutions for monitoring heterogeneity and safeguarding consistent security across various devices.
A. Transfer learning: Transfer learning makes it possible to apply the models developed in one domain to another domain. Such a technique is effective in applying pre-trained security models to new IoT settings with little additional training [31]. It is also helpful in overcoming the difficulties caused by heterogeneous IoT networks through the exploitation of available knowledge.
B. Multi-Task learning: Multi-task learning includes teaching a single model in order to carry out a great number of associated tasks simultaneously. One may use multi-task learning in IoT security to manage a great number of security challenges, e.g., intrusion prevention and anomaly detection, by employing a unified model [32]. Such a technique improves the efficacy and efficiency of security measures.

Real-time detection of threats

To mitigate risk and prevent attacks, it is essential to use real-time threat detection in cybersecurity. Artificial intelligence techniques, especially stream processing and online learning, can be effectively used to evaluate data in real time and respond quickly to threats.
A. Online learning: Online learning algorithms are designed to handle inputs incrementally and modify and update the model upon receiving new data. This technique is particularly effective for IoT systems where threats are constantly changing. In this way, it allows for adaptation to evolving risks and continuous monitoring of the security situation. Through online learning, systems can respond quickly and effectively to threats and identify security issues before a crisis occurs [33].
B. Stream processing: Stream processing frameworks such as Apache Flink and Apache Kafka are powerful tools for analyzing continuous streams of data sent from IoT devices to servers. These networks, especially when combined with artificial intelligence models, provide the ability to detect and monitor security threats in real time [34]. Using such capabilities, the ability to detect and respond to attacks in a timely manner increases, and the necessary preventive measures can be taken. Given the proliferation of cyber threats and their increasing complexity, the use of artificial intelligence methods for real-time threat detection seems essential as a key solution in IoT security. These technologies not only help increase the efficiency of data protection, but can also lead to the creation of intelligent protection systems that remain up-to-date and vigilant.

The use of AI in IoT security poses fundamental privacy and ethical challenges. Protecting user rights and maintaining social trust requires ensuring the proper and effective implementation of AI-based security measures. This requires the development and implementation of comprehensive legal and ethical frameworks that help balance the utility of technology with the individual rights of users.
A. Privacy-preserving methods: Privacy preservation, as one of the fundamental challenges in IoT security, requires innovative and efficient approaches. Techniques such as federated learning and differential privacy are specifically designed to secure sensitive information. In differential privacy, it is ensured that individual data points are not identifiable in real terms, meaning that user information is not disclosed in the analysis process. In addition, federated learning provides an approach to collaboratively train machine learning models in which each user’s raw data is preserved without the need to transfer or disclose it. These methods are especially necessary in sensitive environments such as IoT where a lot of personal data is generated and can help to enhance user privacy [35].
B. Ethical AI: Ethical AI refers to the design and deployment of responsible and transparent AI systems in which biases are minimized and decisions are made based on ethical principles. In the context of IoT security, ethical issues are crucial, including ensuring that user privacy is not violated by security solutions. Also, impartiality and fairness in AI models must be observed to ensure that no group or individual is unfairly deprived of services [36]. Therefore, continuous research and development in this field is crucial to identify and address ethical concerns and build trust in AI-based security solutions. This research will not only improve cybersecurity but also increase awareness and education of users.

Maintaining the security of connected devices is of utmost importance in the ever-changing IoT environment. Sometimes, conventional security measures are not sufficient in IoT settings due to their variety, dynamic nature, and size. The present section suggests a comprehensive, high-level technique to protect IoT devices that incorporates AI-driven judgments. Artificial intelligence can be used to enhance security measures across a variety of domains, from incident response to authentication, and present robust protection against emerging threats. The plan is focused on critical areas in which artificial intelligence may add significant value, e.g., secure communication, device authentication, intrusion detection, frequent updates, privacy protection, risk assessment, continuous development, and incident response (Figure 1).

Figure 1:Strategy for securing IoT devices.

a. Use strong authentication protocols: Use strong authentication protocols to secure IoT devices. By leveraging AI features, it is possible to learn the typical behaviors of each device and identify irregular authentication patterns, such as irregular logins.
b. Secure communications: All data transfers between servers and IoT devices should be fully encrypted. AI can help identify unexpected patterns in network traffic, which are a sign of a security concern and potential threat.
c. Regular updates and patches: IoT devices should be updated regularly to ensure their security. AI can analyze historical data to identify the most vulnerable devices and prioritize them for updates.
d. Intrusion Prevention and Detection: Intrusion Detection Systems (IDS) equipped with AI can detect and respond to attacks in real time. These systems should be able to use machine learning to detect new threats.
e. Risk assessment: Using AI to assess risks based on parameters such as device type, location, activity, etc. can help identify potential vulnerabilities and provide necessary preventive measures.
f. Privacy protection: Effective measures can be taken to protect the privacy of data collected by IoT devices using AI. AI algorithms can be used to anonymize data and remove personally identifiable information (PII).
g. Incident response: A comprehensive strategy should be developed for responding to security incident events. AI can be very effective in analyzing the incident situation and in deciding on the best course of action.
h. Continuous learning and reinforcement: AI models must continually learn from new data collected and refine their predictions and conclusions. This requires a feedback loop in which judgments about AI results are used to improve the model (Figure 1).

For example, for better efficiency, a multilayer hybrid system can be proposed that works as follows:
A. Step 1: Using lightweight techniques such as PCA for initial data processing (including various features such as network traffic, temperature, voltage, device timing, and the like) and identification of obvious anomalies.
B. Step 2: Sending the output data of the first layer to more complex models such as RNN or GAN for deeper pattern analysis and accurate identification of advanced threats.
C. Step 3: Using transfer learning methods as well as edge computing to reduce processing pressure and improve efficiency (Resource optimization).

The topic of AI-based IoT security is rapidly growing and offers a number of interesting topics for future development and study. Topics include: Explainable Artificial Intelligence, federated learning, and quantum machine learning. Explainable Artificial Intelligence (XAI) aims to create AI models that provide understandable and explicit explanations for their actions. This is a critical issue for trust and understanding of AI-based security solutions. XAI research aims to improve the transparency and usefulness of AI in security applications. Federated learning, as a distributed machine learning method, enables collaborative model training while protecting data privacy [37,38]. Future research will focus on improving the scalability and efficiency of federated learning techniques and solving problems related to data connectivity and heterogeneity. Quantum Machine Learning (QML) has the potential to transform AI by extending its computational potential beyond traditional systems and could provide new answers to IoT security concerns, including rapid anomaly detection and pattern recognition. Ongoing research is studying the integration of quantum computing with artificial intelligence to improve security.

The rapid expansion and diversity of Internet of Things (IoT) devices pose significant security challenges due to the increasing attack surface and resource constraints. IoT ecosystems, from lowpower sensors to complex machines, have become increasingly dynamic, which complicates the implementation of traditional security methods. In the meantime, innovative solutions based on advanced technologies, especially artificial intelligence, seem essential to address these challenges. Artificial intelligence-based strategies, such as Reinforcement Learning (RL), Deep Learning (DL), and Explainable Artificial Intelligence (XAI), have opened new horizons for strengthening IoT security. These technologies have the ability to analyze huge amounts of data, identify threats in real time, and optimize security policies to suit dynamic environments. For example, reinforcement learning enables the generation of adaptive security rules, while deep learning is able to detect anomalies and react to complex patterns that indicate potential attacks. In addition, Explainable Artificial Intelligence (XAI) increases user and operator trust by providing transparent and understandable models, paving the way for more informed security decisions. This combination of technologies also enables the use of mathematical models to optimize algorithms, calculate anomaly scores, and predict attack paths based on historical data. These solutions not only reduce the complexity of IoT security, but also create infrastructures that are more resilient to cyber threats. Therefore, the application of these advanced technologies transforms IoT systems into more stable and secure structures, while ensuring the integrity, availability, and confidentiality of information (the CIA triangle). With continuous development and research in this area, we can not only stay ahead of the growing threat landscape, but also be able to effectively deal with emerging security complexities and achieve a more secure future in the IoT.

1. Sicari S, Rizzardi A, Grieco L, Coen-Porisini A (2015) Security, privacy and trust in internet of things: The road ahead. Computer Networks 76: 146-164.
2. Zaman S, Alhazmi K, Aseeri M, Ahmed MR, Khan RT, et al. (2021) Security threats and artificial intelligence-based countermeasures for internet of things networks: A comprehensive survey. IEEE Access 9: 94668-94690.
3. Malik R, Singh Y, Sheikh ZA, Anand P, Singh PK, et al. (2022) An improved deep belief network ids on IoT-based network for traffic systems. Journal of Advanced Transportation 2023(1): 1-18.
4. Bertino E (2016) Data security and privacy in the IoT. EDBT pp. 1-3.
5. Radoglou-Grammatikis PI, Sarigiannidis P, Moscholios I (2019) Securing the internet of things: Challenges, threats and solutions. Internet Things 5: 41-70.
6. Gritti C, Regnath E, Steinhorst S (2021) Practical and scalable access control mechanism for the internet of things using time-bound attribute-based encryption IACR. Cryptol. ePrint Arch 2021: 140.
7. Alaba FA, Othman M, Hashem IAT, Alotaibi F (2017) Internet of things security: A survey. J Netw Comput Appl 88: 10-28.
8. Conti M, Dehghantanha A, Franke K, Watson S (2018) Internet of things security and forensics: Challenges and opportunities. Future Generation Computer Systems 78: 544-546.
9. Xu R, Chen Y, Li X, Blasch E (2022) A secure dynamic edge resource federation architecture for cross-domain IOT systems. 2022 International Conference on Computer Communications and Networks (ICCCN), Honolulu, HI, USA.
10. Zhou J, Cao Z, Dong X, Vasilakos AV (2017) Security and privacy for cloud-based IoT: Challenges. IEEE Communications Magazine 55(1): 26-33.
11. Qu Y, Yu S, Zhou W, Peng S, Wang G, et al. (2018) Privacy of things: Emerging challenges and opportunities in wireless internet of things. IEEE Wireless Communications 25(6): 91-97.
12. Hashemi S, Shams F (2016) Security capabilities and vulnerabilities of internet of things architectures and the necessity of using dynamic security for IoT.
13. Stoyanova M, Nikoloudakis Y, Panagiotakis S, Pallis E, Markakis E (2020) A Survey on the Internet of Things (IoT) forensics: challenges, approaches, and open issues. IEEE Communications Surveys & Tutorials 22(2): 1191-1221.
14. Zia T, Liu P, Han W (2017) Application-specific digital forensics investigative model in Internet of Things (IoT). Proceedings of the 12^th International Conference on Availability, Reliability and Security 55: 1-7.
15. Mishra N, Pandya S (2021) Internet of things applications, security challenges, attacks, intrusion detection, and future visions: A systematic review. IEEE Access 9: 59353-59377.
16. Kanimozhi V, Prem Jacob T (2021) Artificial Intelligence for anomaly detection by employing deep learning strategies in IoT networks using the trendy IoT-23 big data from Google’s tensorflow2.2. Research Square.
17. Rihan SDA, Anbar M, Alabsi BA (2023) Approach for detecting attacks on IOT networks based on ensemble feature selection and deep learning models. Sensors (Basel, Switzerland) 23(17): 7342.
18. Munir M, Siddiqui SA, Dengel A, Ahmed S (2019) DeepAnT: A deep learning approach for unsupervised anomaly detection in time series. IEEE Access 7: 1991-2005.
19. Odeh A, Abu Taleb A (2023) Ensemble-based deep learning models for enhancing IoT intrusion detection. Applied Sciences 13(21): 11985.
20. Mudgerikar A, Bertino E, Lobo J, Verma D (2021) A security-constrained reinforcement learning framework for software defined networks. ICC 2021-IEEE International Conference on Communications, Montreal, Quebec, Canada.
21. Uprety A, Rawat DB (2020) Reinforcement learning for IoT security: A comprehensive survey. IEEE Internet of Things Journal 8(11): 8693-8706.
22. Moerland TM, Broekens J, Plaat A, Jonker CM (2023) Model-based reinforcement learning: A survey. Foundations and Trends® in Machine Learning 16(1): 1-118.
23. Li T, Zhu K, Luong NC, Niyato D, Wu Q, et al. (2021) Applications of multi-agent reinforcement learning in future internet: A comprehensive survey. IEEE Communications Surveys & Tutorials 24(2): 1240 -1279.
24. HaddadPajouh H, Dehghantanha A, Khayami R, Choo KKR (2018) A deep recurrent neural network-based approach for internet of things malware threat hunting. Future Generation Computer Systems 85: 88-96.
25. T Li, Sahu AK, Talwalkar Ameet, Smith V (2020) Federated learning: Challenges, methods, and future directions. IEEE Signal Processing Magazine 37(3): 50-60.
26. Wang C, Wang B, Liu H, Qu H (2020) Anomaly detection for industrial control system based on autoencoder neural network. Wireless Communications and Mobile Computing 2020: 8897926:1-8897926:10.
27. Ong KSH, Wang W, Hieu NQ, Niyato D, Friedrichs T (2022) Predictive maintenance model for IIoT-based manufacturing: A transferable deep reinforcement learning approach. IEEE Internet of Things Journal 9(17): 15725-15741.
28. Bikmukhamedov RF, Nadeev AF (2019) Lightweight machine learning classifiers of IoT traffic flows. In 2019 Systems of Signal Synchronization, Generating and Processing in Telecommunications (SYNCHROINFO), Russia, pp. 1-5.
29. Chen J, Chen S, Wang Q, Cao B, Feng G, et al. (2019) iRAF: A deep reinforcement learning approach for collaborative mobile edge computing IOT networks. IEEE Internet of Things Journal 6(6): 7011-7024.
30. Upadhyaya P, Ruchi, Dutt S (2022) Transfer learning approach for 6G-IoT applications. 2022 7^th International Conference on Communication and Electronics Systems (ICCES), Coimbatore, India, pp. 421-424.
31. Dong H, Kotenko I (2023) An autoencoder-based multi-task learning for intrusion detection in IOT networks. 2023 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT), Yekaterinburg, Russia, pp. 1-4.
32. Li FQ, Zhao RJ, Wang SL, Chen LB, Liew AWC, et al. (2022) Online intrusion detection for internet of things systems with full bayesian possibilistic clustering and ensembled fuzzy classifiers. IEEE Transactions on Fuzzy Systems 30(11): 4605-4617.
33. Barry M, El Jaouhari S, Bifet A, Montiel J, Guerizec E (2022) StreamFlow: A system for summarizing and learning over industrial big data streams. 2022 IEEE International Conference on Big Data (Big Data), Osaka, Japan, pp. 2198-2205.
34. Ruzafa-Alcazar P, Fernandez-Saura P, Marmol-Campos E, González-Vidal A, Hernández-Ramos JL, et al. (2023) Intrusion detection based on privacy-preserving federated learning for the industrial IoT. IEEE Transactions on Industrial Informatics 19(2): 1145-1154.
35. Gupta N (2023) Artificial intelligence ethics and fairness: A study to address bias and fairness issues in AI systems, and the ethical implications of AI applications. Revista Review Index Journal of Multidisciplinary 3(2): 24-35.
36. Jiang F, Fu Y, Gupta B, Liang Y, Rho S, et al. (2020) Deep learning based multi-channel intelligent attack detection for data security. IEEE Transactions on Sustainable Computing 5(2): 204-212.
37. Oseni A, Moustafa N, Creech G, Sohrabi N, Strelzoff A, et al. (2023) An explainable deep learning framework for resilient intrusion detection in IoT-enabled transportation networks. IEEE Transactions on Intelligent Transportation Systems 24(1): 1000-1014.
38. Benaddi H, Jouhari M, Ibrahimi K, Benslimane A, Amhoud EM (2022) Adversarial attacks against IoT networks using conditional GAN based learning. 2022 IEEE Global Communications Conference (GLOBECOM), Brazil, pp. 2788-2793.