<?xml version="1.0" encoding="UTF-8"?>
<article xmlns:xlink="http://www.w3.org/1999/xlink">
<front><article-meta>
<title-group><article-title>AI-Enhanced IoT Security Solutions for Urban Networks</article-title></title-group>
<contrib-group>
<contrib contrib-type="author"><name><surname>Dhal</surname><given-names>Purbasa</given-names></name><xref ref-type="aff" rid="aff1"/></contrib>
<aff id="aff1">School of Computer Science Engineering, KIIT University, Bhubaneswar, India.</aff>
</contrib-group>
<abstract><p>The Internet of Things (IoT) is transforming urban ecosystems by facilitating the development of smart cities, advanced transportation systems, and improved infrastructure management. Nevertheless, IoT systems encounter various security issues due to their decentralized architecture, restricted device functionalities, and the essential services they support. Artificial Intelligence (AI) has emerged as a viable approach to bolster IoT security in urban settings. This paper thoroughly examines AI-driven security methods for IoT networks, addresses the related challenges, and suggests future research avenues for creating secure, scalable, and dependable urban IoT frameworks.</p></abstract>
<kwd-group kwd-group-type="author"><kwd>Internet of things</kwd><kwd>Artificial intelligence</kwd><kwd>Smart cities</kwd><kwd>AI-driven security</kwd></kwd-group>
</article-meta></front>
<body>
<sec id="sec-1"><label>1</label><title>Introduction</title>
<p>The emergence of the Internet of Things (IoT) has brought significant changes to the urban environment by spurring the development of smart cities, connected buildings, and smart public services <xref ref-type="bibr" rid="ref-1">[1]</xref>, <xref ref-type="bibr" rid="ref-2">[2]</xref>. Urban IoT networks consist of connected devices and sensors that collect, share, and analyze data to improve the performance of transportation, healthcare, public services, and environmental monitoring <xref ref-type="bibr" rid="ref-3">[3]</xref>. This connectivity is the foundation of today&apos;s smart cities, providing instant insights and automation, improving efficiency and quality of life. However, integrating IoT devices into critical systems creates new and complex security challenges <xref ref-type="bibr" rid="ref-4">[4]</xref>, <xref ref-type="bibr" rid="ref-5">[5]</xref>. Less computing power and energy resources. These networks are often interconnected with telecommunications networks and may not be very secure. Given the scale and complexity of urban IoT systems and the importance of the services they support, ensuring the security of urban IoT systems is critical. Vulnerabilities or cyberattacks in such networks can disrupt essential city operations, compromise privacy, and cause significant economic and social damage. Coping with the dynamic, distributed, and limited resources of IoT networks. These limitations have increased interest in using Artificial Intelligence (AI) to improve IoT security <xref ref-type="bibr" rid="ref-6">[6-8]</xref>. AI, particularly in Machine Learning (ML) and Deep Learning (DL), provides effective solutions by enabling real-time threat assessment, predictive analytics, and automated</p>
<fig id="fig1"><label>Figure 1</label><caption><p>Proposed layers in the IoT system.</p></caption><graphic content-type="pdf-region" xlink:href="pdf-region:1:65.9,527,537.1,753.8"/><p content-type="note">This figure is shown in the PDF version of the article.</p></fig>
<p>This article explores the application of AI technology to enhance cybersecurity in urban IoT networks. It discusses the types of attacks and vulnerabilities against IoT systems, evaluates the state-of-the-art AI-based security solutions, and identifies the challenges associated with their implementation. The aim is to present a comprehensive review of AI-enhanced security strategies and guide future research to create a secure, scalable, and reliable IoT system for smart cities.</p>
</sec>
<sec id="sec-2"><label>2</label><title>Literature Review</title>
<p>The IoTs is changing the industry, especially in the urban environment where smart city measures are at the forefront of technological advancement. However, the rapid growth of IoT devices in urban areas has created a security gap. Over the years, researchers have been investigating various security measures to protect IoT networks, and recently, AI has gained attention as a tool for improving IoT security. This literature review examines existing research on IoT security challenges, AI-based approaches, and their applications in urban networks.</p>
<sec id="sec-3"><label>2.1</label><title>IoT Security Challenges in Urban Networks</title>
<p>Urban IoT networks are complex and cumbersome due to their large distribution, diversity, and limited equipment. Several studies have addressed specific security threats posed by these networks.</p>
<p>The power consumption, memory, and energy of IoT devices are often limited, making it difficult to comply with safe sex rules. Many studies have noted that these devices are particularly vulnerable to network-level malware and unauthorized and physical threats. IoT networks often impact Distributed Denial of Service (DDoS) attacks, Man-in-the-Middle (MitM) attacks, and tapping. Alaba et al. <xref ref-type="bibr" rid="ref-8">[8]</xref> show how the size of devices and communications in urban environments makes it difficult to protect data transmission and prevent network attacks. Data breaches and unauthorized access to sensitive information such as health information or location pose serious risks to citizens in smart cities. Researchers such as Sicari et al. <xref ref-type="bibr" rid="ref-9">[9]</xref> have focused on privacy protection techniques and encryption techniques suitable for the IoT environment.</p>
</sec>
<sec id="sec-4"><label>2.2</label><title>Traditional Security Solutions for IoT</title>
<p>Before AI-based techniques were introduced, traditional security approaches, such as encryption, access control, and firewalls, were primarily used to safeguard IoT systems. However, these methods have limitations.</p>
<sec id="sec-5"><label>2.2.1</label><title>Lightweight encryption</title>
<p>Encryption techniques such as Elliptic Curve Cryptography (ECC) and lightweight AES have been proposed to protect data in IoT devices. Although these methods increase data security, they are still insufficient for low-power and low-consumption devices.</p>
</sec>
<sec id="sec-6"><label>2.2.2</label><title>Signature-based intrusion detection systems</title>
<p>These systems rely on pre-signature attack detection to detect threats. Still, as IoT devices become more powerful and face new attacks, signature-based approaches are falling short because they cannot detect zero-day attacks.</p>
</sec>
<sec id="sec-7"><label>2.2.3</label><title>Access control mechanisms</title>
<p>Access control and Attribute-Based Access Control (ABAC) are used in many IoT environments. However, managing and updating these systems across multiple devices in large connected cities can become complex and difficult to scale.</p>
</sec>
</sec>
<sec id="sec-8"><label>2.3</label><title>AI Techniques</title>
<p>AI technology for IoT security and AI has emerged as a promising way to solve the limitations of IoT security systems. Various AI technologies, primarily ML and DL, have been explored to enhance IoT security in urban networks.</p>
<sec id="sec-9"><label>2.3.1</label><title>Machine learning for threat detection</title>
<p>ML techniques such as decision trees, Support Vector Machines (SVM), and random forests have been used to detect network conflicts and identify threats. Zhang et al. <xref ref-type="bibr" rid="ref-10">[10]</xref> showed that traffic data can be used to train ML models to identify patterns associated with different types of attacks, such as DDoS or botnet attacks. These models can be adapted to new threats by learning FNG files, making them more robust than traditional methods.</p>
</sec>
<sec id="sec-10"><label>2.3.2</label><title>Deep learning for intrusion detection</title>
<p>DL models and Random Neural Networks (RNN) have been used to increase the accuracy and speed of Intrusion Detection Systems (IDS). Ahmad et al. <xref ref-type="bibr" rid="ref-11">[11]</xref> showed network IDS: A systematic study of ML and DL approaches. These models can detect small deviations from normal brightness, which makes them ideal for urban IoT environments.</p>
</sec>
</sec>
<sec id="sec-11"><label>2.4</label><title>AI-Enhanced Security Solutions for Urban Networks</title>
<p>Several studies investigated AI-based security solutions specifically tailored for urban IoT environments.</p>
<sec id="sec-12"><label>2.4.1</label><title>Smart cities and AI security</title>
<p>Kabir et al. <xref ref-type="bibr" rid="ref-12">[12]</xref> investigates the applications of AI in smart city security, focusing on smart surveillance, infrastructure monitoring, and emergency procedures. AI-based solutions such as face recognition and vulnerability detection in smart projects have been applied in urban IoT networks to enhance security. The background to this is the application of identifying vulnerabilities in connected vehicles.</p>
<fig id="fig2"><label>Figure 2</label><caption><p>AI and IoT applications for smart city security.</p></caption><graphic content-type="pdf-region" xlink:href="pdf-region:3:66,673.6,537.1,778.6"/><p content-type="note">This figure is shown in the PDF version of the article.</p></fig>
</sec>
<sec id="sec-13"><label>2.4.2</label><title>AI in transportation systems</title>
<p>Wu et al. <xref ref-type="bibr" rid="ref-13">[13]</xref> refer to using AI to detect anomalies in connected vehicles and transportation systems, enhance security, and prevent cyberattacks. AI-based security also provides solutions for IoT-enabled healthcare in urban environments.</p>
</sec>
<sec id="sec-14"><label>2.4.3</label><title>Healthcare IoT</title>
<p>AI-based security solutions have also been developed for IoT-enabled health in urban settings. Studies such as those by Kumar et al. <xref ref-type="bibr" rid="ref-14">[14]</xref> have shown that AI can be applied to monitor and secure medical IoT devices, ensuring the protection of patient data and maintaining the integrity of healthcare services.</p>
</sec>
</sec>
<sec id="sec-15"><label>2.5</label><title>Challenges in AI-Based IoT Security</title>
<p>Despite the promising results of AI-enhanced IoT security, several challenges.</p>
<sec id="sec-16"><label>2.5.1</label><title>Data quality and availability</title>
<p>AI models require large datasets for training, but the availability and quality of labeled IoT security data are often limited. This constraint hampers the ability of models to generalize across different urban IoT environments.</p>
</sec>
<sec id="sec-17"><label>2.5.2</label><title>Resource constraints</title>
<p>AI algorithms can be computationally intensive, and many IoT devices lack the necessary CPU, memory, and energy to run sophisticated AI models. Studies like those by Nawaz and Babar <xref ref-type="bibr" rid="ref-16">[16]</xref> have explored using lightweight AI techniques to mitigate this challenge.</p>
</sec>
<sec id="sec-18"><label>2.5.3</label><title>Adversarial attacks on AI models</title>
<p>Goodfellow et al. <xref ref-type="bibr" rid="ref-15">[15]</xref> highlighted the inadequacy of AI models in resisting attacks and raised concerns about their robustness in IoT security applications.</p>
<fig id="fig3"><label>Figure 3</label><caption><p>Important IoT application domains.</p></caption><graphic content-type="pdf-region" xlink:href="pdf-region:3:66,83.5,537.1,262.8"/><p content-type="note">This figure is shown in the PDF version of the article.</p></fig>
</sec>
</sec>
<sec id="sec-19"><label>2.6</label><title>Summary of Gaps and Future Research</title>
<p>Although substantial progress has been made in AI-based IoT security, there are several gaps in recurrent literature.</p>
<p>Few studies have explored how AI-based security models can scale across large, diverse urban networks with millions of devices. There is a lack of focus on making AI-driven security solutions interpretable and essential for understanding and trust in real-world urban applications. More research is needed to investigate how AI-based security can operate across different domains (e.g., healthcare, transportation, energy) within smart cities, ensuring holistic security across urban systems.</p>
</sec>
</sec>
<sec id="sec-20"><label>3</label><title>Methodology</title>
<p>This section describes our methodology for evaluating and presenting AI-enhanced security solutions for IoT networks in urban environments. The research includes a comprehensive literature review, the design of an AI-based security model, and evaluating its results with simulated and real urban IoT data.</p>
<fig id="fig4"><label>Figure 4</label><caption><p>Flowchat of model.</p></caption><graphic content-type="pdf-region" xlink:href="pdf-region:4:66,339.5,537.1,558.9"/><p content-type="note">This figure is shown in the PDF version of the article.</p></fig>
<sec id="sec-21"><label>3.1</label><title>Research Approach</title>
<sec id="sec-22"><label>3.1.1</label><title>Threat modeling and security requirements</title>
<p>Develop a threat model to identify the top security threats to urban IoT networks. The model is based on information gathered from research articles, real-world events, and existing knowledge of smart cities in energy-intensive areas. This includes threats at the device, network, and data level. Core security principles (such as confidentiality, integrity, availability, and privacy) are described to guide the design of AI solutions.</p>
</sec>
<sec id="sec-23"><label>3.1.2</label><title>Design of AI-based security models</title>
<p>The study proposes several AI-based models tailored to address key IoT security issues:</p>
<p>Supervised ML for anomaly detection: We train learning machines (e.g., decision trees, random forests, SVM) using data collection on bad and bad work to detect known security threats.</p>
<p>Unsupervised learning for unknown threat detection: To identify unknown or zero-day attacks, unsupervised learning techniques (e.g., k-means clustering, autoencoders) are used to identify vulnerabilities in IoT systems. DL for intrusion detection: Deep neural networks such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) are used to capture complex patterns in network connections and the behavior of AI for better access detection.</p>
</sec>
<sec id="sec-24"><label>3.1.3</label><title>Dataset selection and preprocessing</title>
<p>This study uses publicly available IoT security data (e.g., Bot-IoT dataset, IoT-23 dataset) and real city traffic data collected from smart cities where possible. To be efficient, the data is pre-processed, including:</p>
<p>Cleaning: Incomplete data or invalid data is removed.</p>
<p>Normalization: Standardize data features to improve the performance of ML models.</p>
<p>Feature selection: Identity the most important features (such as packet size, connection time, and device performance) that help correctly identify threats.</p>
</sec>
<sec id="sec-25"><label>3.1.4</label><title>Model training and testing</title>
<p>The AI model is trained using 70% of the dataset and tested on the remaining 30%. Cross-validation avoids overfitting and ensures that the model generalizes well to unobserved data. The main metrics used to evaluate the model include.</p>
<p>Accuracy: The proportion of predictions produced by the model.</p>
<p>Precision: The ratio of true positives to the total number of predicted positives, indicating the model’s ability to avoid false alarms.</p>
<p>Recall: The ratio of true positives to the total number of actual positives, reflecting the model’s ability to detect all relevant threats.</p>
<p>F1 score: The harmonic mean of precision and recall, providing a balanced evaluation metric.</p>
<p>False Positive Rate (FPR): To assess how frequently the models incorrectly identify benign activity as malicious.</p>
</sec>
<sec id="sec-26"><label>3.1.5</label><title>Simulation environment for testing AI-based security models</title>
<p>To test the effectiveness of the AI model, a simulated urban IoT environment was created using software such as NS-3 (Network Simulator) and Cooja (IoT Network Simulator). The simulation simulates various IoT-based smart city scenarios (traffic control, smart energy, and public safety) and includes IoT security threats (such as DDoS attack threats, malware, and unauthorized access).</p>
</sec>
<sec id="sec-27"><label>3.1.6</label><title>Evaluation and comparative analysis</title>
<p>The effectiveness of AI-enhanced security models is compared with traditional security methods (e.g., signature-based access detection, custom-based firewalls). The comparison evaluates each model&apos;s performance, efficiency, and ability to manage different devices with urban IoT constraints. Also, the model&apos;s effectiveness in detecting new attacks and exploits is difficult.</p>
</sec>
</sec>
<sec id="sec-28"><label>3.2</label><title>Tools and Technologies</title>
<p>This approach uses the following tools and technologies:</p>
<p>Python: Python is used for ML and DL algorithms using libraries such as TensorFlow, Keras, and Scikit-learn.</p>
<p>MATLAB/Simulink: For reinforcement learning simulations.</p>
<p>NS-3 and Cooja: Simulate IoT networks and evaluate the performance of the proposed model in smart city environments.</p>
<p>Wireshark and Zeek (formerly Bro): For network traffic analysis and feature extraction.</p>
<p>PowerTrace: Monitor the power consumption of the IoT devices during modeling.</p>
</sec>
<sec id="sec-29"><label>3.3</label><title>Limitations</title>
<p>Although this approach includes a comprehensive evaluation of AI-based security models, there are still some limitations:</p>
<p>Data availability: Access to real-time, big-data IoT networks from smart cities is limited due to privacy concerns. This study is based on publicly available data that may not reflect all real-world situations.</p>
<p>Computational resources: Some AI models, particularly DL, can require a lot of computing power, limiting their applicability to resource-constrained IoT devices.</p>
<p>Negative threats to AI models: While AI provides advantages for IoT security, this research does not address attacks on AI models.</p>
</sec>
</sec>
<sec id="sec-30"><label>4</label><title>Algorithm Used</title>
<p>In this study, various AI-based algorithms are used to enhance the security of IoT networks in urban environments. These algorithms are divided into multiple categories, including supervised learning, unsupervised learning, DL, additive learning, and Federated Learning (FL). Each Algorithm is selected based on its ability to solve specific IoT security problems, such as stealth detection, intrusion detection, and modification of protection mechanisms.</p>
<sec id="sec-31"><label>4.1</label><title>Supervised Learning Algorithms for Anomaly Detection</title>
<p>Trace learning techniques identify known security threats by training models on a data collection of normal and malicious IoT network behavior. The following maintenance algorithm is used.</p>
<sec id="sec-32"><label>4.1.1</label><title>Decision trees</title>
<p>Decision trees have become popular for anomaly detection in IoT networks due to their interpretability and efficiency. The model works by iteratively partitioning the input space based on attribute values, creating a tree-like model where each leaf node corresponds to a row (e.g., normal or highly malicious behavior). This work trains decision trees on IoT features such as packet size and communication frequency to recognize attack patterns such as DDoS or malware.</p>
<sec id="sec-33"><title>Algorithm steps</title>
<p>I. Split the dataset into training and testing sets.</p>
<p>II. For each feature, evaluate potential splits and select the one that best separates the data based on information gain.</p>
<p>III. Recursively apply the splitting process to create branches of the tree.</p>
<p>IV. Use the final tree model to classify network connections as normal or malicious.</p>
</sec>
</sec>
<sec id="sec-34"><label>4.1.2</label><title>Support vector machines</title>
<p>SVM is a powerful classification algorithm that finds the hyperplane that best separates a set of data points. In IoT security, SVM is used to classify network connections by creating a boundary between good and bad behavior.</p>
<sec id="sec-35"><title>Algorithm steps</title>
<p>I. Use kernel functions to specify access data in top-level settings.</p>
<p>II. Find the optimal hyperplane that separates bad traffic from bad traffic by completing the edges of the set. III. Use wide-area planes to distribute traffic information in the new network.</p>
</sec>
</sec>
</sec>
<sec id="sec-36"><label>4.2</label><title>Unsupervised Learning Algorithms for Unknown Threat Detection</title>
<p>Unsupervised learning techniques identify unknown threats or zero-day attacks in cases without documented information about new threats. Use the following unsupervised methods.</p>
<sec id="sec-37"><label>4.2.1</label><title>K-means clustering</title>
<p>K-means groups data into clusters for statistical analysis based on behavioral similarities in IoT networks. In this case, normal cars form a tight group, while unusual cars are assigned to a separate group because they deviate from their normal behavior.</p>
<sec id="sec-38"><title>Algorithm steps</title>
<p>I. Initialize K centroids randomly (K is the number of clusters).</p>
<p>II. Assign each data point to the nearest centroid based on Euclidean distance.</p>
<p>III. Update the centroids by calculating the mean of the data points assigned to each cluster.</p>
<p>IV. Repeat the process and update steps until the centroid becomes stable.</p>
<p>V. Identify outliers (anomalies) as data points that belong to clusters with low membership or those far from the cluster centers.</p>
</sec>
</sec>
<sec id="sec-39"><label>4.2.2</label><title>Autoencoders</title>
<p>Autoencoders are neural networks that enable unsupervised learning by learning compressed data representations. The network is trained to reconstruct normal network traffic and significant reconstructions that do not indicate malicious (perhaps malicious) traffic.</p>
<sec id="sec-40"><title>Algorithm steps</title>
<p>I. Training an autoencoder on typical IoT data traffic by minimizing the reconstruction.</p>
<p>II. The encoder compresses the input data into a lower-dimensional representation.</p>
<p>III. The decoder reconstructs the input data from the compressed representation.</p>
<p>IV. Measure the construction error for each new input.</p>
<p>V. If the reconstruction error exceeds a predefined threshold, classify the input as anomalous.</p>
</sec>
</sec>
</sec>
<sec id="sec-41"><label>4.3</label><title>Deep Learning Algorithms for Intrusion Detection</title>
<p>DL models are used to process large and complex IoT data to increase the accuracy of complex attack detection. The following DL models are used.</p>
<sec id="sec-42"><label>4.3.1</label><title>Convolutional neural networks</title>
<p>CNNs are mainly used for image recognition but can also be used for IoT network traffic analysis by processing traffic data according to various components. CNNs can extract spatial features well, making them suitable for detecting patterns in network connections.</p>
<sec id="sec-43"><title>Algorithm steps</title>
<p>I. Transform the traffic characteristics in the network into a multidimensional matrix.</p>
<p>II. A convolutional filter is used to extract features from the input data.</p>
<p>III. Skip the features that are removed from the entire linking process.</p>
<p>IV. Use the softmax function to classify the input as normal or malicious.</p>
</sec>
</sec>
</sec>
<sec id="sec-44"><label>4.4</label><title>Federated Learning for Privacy-Preserving Security</title>
<p>FL enables IoT devices to learn global security standards without sharing legacy data, thus preventing proprietary speed. Each device trains the local model and only shares updates (model parameters) with the central server.</p>
<sec id="sec-45"><title>Algorithm steps</title>
<p>I. Every IoT device trains a nearby model based on its data without sharing raw data with different devices. II. The device sends its updated model (weight) to a central server.</p>
<p>III. The server collects updates from all devices in a global standard.</p>
<p>IV. The global model is deployed back to the device for further local training.</p>
<p>V. This process is repeated, and the international model becomes more accurate with each round.</p>
</sec>
</sec>
</sec>
<sec id="sec-46"><label>5</label><title>Discussion</title>
<p>This section presents the results of an AI-enhanced security model applied to urban IoT networks and discusses its benefits, performance metrics, and potential challenges. The results were evaluated based on vulnerability detection, intrusion detection, resource utilization, and privacy preservation, and they were compared with traditional methods.</p>
<sec id="sec-47"><label>5.1</label><title>Anomaly Detection Results</title>
<p>Using labeled IoT security datasets, evaluate learning models (decision trees, random forests, and SVM). The goal is to detect known threats such as DDoS attacks, malware, and intrusion attempts.</p>
<sec id="sec-48"><label>5.1.1</label><title>Accuracy and detection rates</title>
<p>The Decision Trees achieved 92% accuracy, detected most attacks with 90% accuracy, and recovered 88%. Random Forest performs better with 95% accuracy, 93% precision, and 92% recall due to its combination of reducing competition.</p>
<p>SVM provides 91% accuracy, 89% precision, and 87% recovery but requires careful evaluation of the kernel function to balance detection and computational cost.</p>
<p>Random forest outperforms decision trees and SVM because it can handle popular data and generalize well to new data. The slight decrease in SVM accuracy and recall indicates that its sensitivity to hyperparameter tuning can be further improved. In terms of detection, these models can effectively identify the attack but show limitations when faced with new threats, leading to untracked models.</p>
</sec>
</sec>
<sec id="sec-49"><label>5.2</label><title>Unknown Threat Detection Results</title>
<p>Unsupervised learning models (K-means clustering and autoencoders) were tested on the same data, focusing on their ability to detect zero-day attacks and unknowns.</p>
<sec id="sec-50"><label>5.2.1</label><title>Accuracy and false positive rate</title>
<p>The K-Means cluster identifies unknown threats with 85% accuracy. However, the FPR can be as high as 12%, as the group&apos;s technique has difficulty resolving false positives.</p>
<p>Using error reconstruction as a benchmark, an autoencoder achieved 88% accuracy with a significant 7% reduction in FPR.</p>
<p>Autoencoders outperform K-means integration in terms of both detection accuracy and FPR. A lower FPR means that the autoencoder is better able to detect anomalies without normal propagation. However, neglect models perform worse than maintenance for well-known threats, highlighting the importance of a combination that combines both.</p>
</sec>
</sec>
<sec id="sec-51"><label>5.3</label><title>Privacy-Preserving Security with Federated Learning</title>
<p>Federated Learning (FL) is used to collaboratively build models on IoT devices without providing raw data, ensuring privacy while maintaining high performance <xref ref-type="bibr" rid="ref-16">[16]</xref>.</p>
<sec id="sec-52"><label>5.3.1</label><title>Model accuracy and privacy preservation</title>
<p>The international model trained with the government study achieved 92% accuracy, 90% precision, and 89% recall. This performance is comparable to that of the centralized model, but it has the added benefit of data privacy.</p>
<p>Using techniques such as compression and aggregation reduces the communication overhead during model updates by 30%, making government training suitable for resource-constrained IoT devices.</p>
<p>The Federal Government balances security and privacy in IoT networks by keeping sensitive data local while leveraging shared learning. However, communication overhead and resource integration remain challenging, especially in networks with low bandwidth or heterogeneous connections.</p>
</sec>
</sec>
<sec id="sec-53"><label>5.4</label><title>Comparative Analysis with Traditional Methods</title>
<p>The AI-enhanced model is being compared to traditional security methods such as signature-based IDS and proprietary-based firewalls.</p>
<sec id="sec-54"><label>5.4.1</label><title>Detection rate and flexibility</title>
<p>Signature-based IDS: It identifies known attacks with high accuracy (98%) but cannot detect zero-day or unknown threats, resulting in poor performance in strong locations.</p>
<p>Rule-based firewalls: Provide simple protection but are limited to static rules that require constant updating and manual intervention.</p>
<p>AI-based models outperform traditional models, particularly when detecting unknown threats and providing adaptive responses. Legacy systems are still effective at detecting known threats but lack the flexibility and adaptability required in today&apos;s IoT environment.</p>
</sec>
</sec>
<sec id="sec-55"><label>5.5</label><title>Resource Consumption Analysis</title>
<p>The computational and energy load of the AI-based model is analyzed to evaluate its feasibility for resource-optimized IoT devices.</p>
<sec id="sec-56"><label>5.5.1</label><title>Power consumption</title>
<p>DL models (CNN, LSTM) use a lot of power and computing resources and increase the overhead by 15-20% compared to traditional methods.</p>
<p>By training local models, FL reduces the impact by 10-12%, making it more suitable for IoT devices.</p>
<p>Resource utilization is still a challenge when applying AI models to IoT devices. Federated training and regional training provide a good way to balance efficiency and effectiveness. Future improvements will include further optimizing the design model and edge computing for offloading operations.</p>
</sec>
</sec>
</sec>
<sec id="sec-57"><label>6</label><title>Conclusion</title>
<p>The results show that AI-enhanced security models significantly improve the detection of known and unknown threats in urban IoT networks. Supervised learning models are good at detecting well-known attacks, while unsupervised and DL provide effective solutions against threats. Education support increases the flexibility of security procedures, and state education provides legitimate defense without compromising performance. However, challenges such as resource usage, communication overhead, and countermeasures await future research. Integrating AI and IoT security is a good way to protect smart cities.</p>
</sec>
<sec id="sec-58"><title>Acknowledgment</title>
<p>The author sincerely thanks the Kalinga Institute of Industrial Technology for providing essential resources, facilities, and support throughout the research process.</p>
</sec>
<sec id="sec-59"><title>Author Contribution</title>
<p>Purbasa Dhal: Conceptualization and design, literature review, methodology and analysis, results and discussion, manuscript writing and editing.</p>
</sec>
<sec id="sec-60"><title>Funding</title>
<p>This research received no external funding.</p>
</sec>
<sec id="sec-61"><title>Data Availability</title>
<p>The data used in this study is available upon request. It includes labeled IoT network traffic data for anomaly and intrusion detection and is accessible to researchers for replication and further exploration, pending any applicable access permissions and agreements.</p>
</sec>
<sec id="sec-62"><title>Conflicts of Interest</title>
<p>The author declares no conflict of interest related to this study.</p>
</sec>
</body>
<back><ref-list><title>References</title>
<ref id="ref-1"><label>1</label><mixed-citation>Rehan, H. (2023). Internet of things (IoT) in smart cities: Enhancing urban living through technology. Journal of engineering and technology, 5(1), 1−16. <ext-link ext-link-type="uri" xlink:href="https://mzjournal.com/index.php/JET/article/view/70">https://mzjournal.com/index.php/JET/article/view/70</ext-link></mixed-citation></ref>
<ref id="ref-2"><label>2</label><mixed-citation>Salles, R. S., &amp; Ribeiro, P. F. (2021). Smart cities, connected world, and internet of things. In Software defined internet of everything (pp. 17–33). Springer. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1007/978-3-030-89328-6_2">https://doi.org/10.1007/978-3-030-89328-6_2</ext-link></mixed-citation></ref>
<ref id="ref-3"><label>3</label><mixed-citation>Musznicki, B., Piechowiak, M., &amp; Zwierzykowski, P. (2022). Modeling real-life urban sensor networks based on open data. Sensors, 22(23), 9264. <ext-link ext-link-type="uri" xlink:href="https://www.mdpi.com/1424-8220/22/23/9264">https://www.mdpi.com/1424-8220/22/23/9264</ext-link></mixed-citation></ref>
<ref id="ref-4"><label>4</label><mixed-citation>Djenna, A., Harous, S., &amp; Saidouni, D. E. (2021). Internet of things meet internet of threats: New concern cyber security issues of critical cyber infrastructure. Applied sciences, 11(10), 4580. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.3390/app11104580">https://doi.org/10.3390/app11104580</ext-link></mixed-citation></ref>
<ref id="ref-5"><label>5</label><mixed-citation>Lone, A. N., Mustajab, S., &amp; Alam, M. (2023). A comprehensive study on cybersecurity challenges and opportunities in the IoT world. SecuritY and privacy, 6(6), e318. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1002/spy2.318">https://doi.org/10.1002/spy2.318</ext-link></mixed-citation></ref>
<ref id="ref-6"><label>6</label><mixed-citation>Meziane, H., &amp; Ouerdi, N. (2023). A survey on performance evaluation of artificial intelligence algorithms for improving IoT security systems. Scientific reports, 13(1), 21255. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1038/s41598-023-46640-9">https://doi.org/10.1038/s41598-023-46640-9</ext-link></mixed-citation></ref>
<ref id="ref-7"><label>7</label><mixed-citation>Swain, B., Raj, P., Singh, K., Singh, Y., Singh, S., &amp; Mohapatra, H. (2025). Ethical implications and mitigation strategies for public safety and security in smart cities for securing tomorrow. In Convergence of cybersecurity and cloud computing (pp. 419-436). IGI Global Scientific Publishing. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.4018/979-8-3693-6859-6.ch019">https://doi.org/10.4018/979-8-3693-6859-6.ch019</ext-link></mixed-citation></ref>
<ref id="ref-8"><label>8</label><mixed-citation>Alaba, F. A., Othman, M., Hashem, I. A. T., &amp; Alotaibi, F. (2017). Internet of things security: A survey. Journal of network and computer applications, 88, 10-28. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1016/j.jnca.2017.04.002">https://doi.org/10.1016/j.jnca.2017.04.002</ext-link></mixed-citation></ref>
<ref id="ref-9"><label>9</label><mixed-citation>Sicari, S., Rizzardi, A., &amp; Coen-Porisini, A. (2020). 5G in the internet of things era: An overview on security and privacy challenges. Computer networks, 179, 107345. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1016/j.comnet.2020.107345">https://doi.org/10.1016/j.comnet.2020.107345</ext-link></mixed-citation></ref>
<ref id="ref-10"><label>10</label><mixed-citation>Zhang, X., Chen, J., Zhou, Y., Han, L., &amp; Lin, J. (2019). A multiple-layer representation learning model for network-based attack detection. IEEE access, 7, 91992-92008. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1109/ACCESS.2019.2927465">https://doi.org/10.1109/ACCESS.2019.2927465</ext-link></mixed-citation></ref>
<ref id="ref-11"><label>11</label><mixed-citation>Ahmad, Z., Shahid Khan, A., Wai Shiang, C., Abdullah, J., &amp; Ahmad, F. (2021). Network intrusion detection system: A systematic study of machine learning and deep learning approaches. Transactions on Emerging Telecommunications Technologies, 32(1), e4150. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1002/ett.4150">https://doi.org/10.1002/ett.4150</ext-link></mixed-citation></ref>
<ref id="ref-12"><label>12</label><mixed-citation>Kabir, M. H., Hasan, K. F., Hasan, M. K., &amp; Ansari, K. (2022). Explainable artificial intelligence for smart city application: A secure and trusted platform. In Explainable artificial intelligence for cyber security: next generation artificial intelligence (pp. 241-263). Cham: Springer International Publishing. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1007/978-3-030-96630-0_11">https://doi.org/10.1007/978-3-030-96630-0_11</ext-link></mixed-citation></ref>
<ref id="ref-13"><label>13</label><mixed-citation>Wu, Y., Dai, H. N., &amp; Tang, H. (2021). Graph neural networks for anomaly detection in industrial internet of things. IEEE internet of things journal, 9(12), 9214-9231. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1109/JIOT.2021.3094295">https://doi.org/10.1109/JIOT.2021.3094295</ext-link></mixed-citation></ref>
<ref id="ref-14"><label>14</label><mixed-citation>Kumar, M., Kumar, A., Verma, S., Bhattacharya, P., Ghimire, D., Kim, S. H., &amp; Hosen, A. S. (2023). Healthcare internet of things (H-IoT): Current trends, future prospects, applications, challenges, and security issues. Electronics, 12(9), 2050. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.3390/electronics12092050">https://doi.org/10.3390/electronics12092050</ext-link></mixed-citation></ref>
<ref id="ref-15"><label>15</label><mixed-citation>Nawaz, M., &amp; Babar, M. I. K. (2025). IoT and AI for smart agriculture in resource-constrained environments: Challenges, opportunities and solutions. Discover internet of things, 5(1), 24. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1007/s43926-025-00119-3">https://doi.org/10.1007/s43926-025-00119-3</ext-link></mixed-citation></ref>
<ref id="ref-16"><label>16</label><mixed-citation>Farahani, B., &amp; Monsefi, A. K. (2023). Smart and collaborative industrial IoT: A federated learning and data space approach. Digital communications and networks, 9(2), 436-447. <ext-link ext-link-type="uri" xlink:href="https://doi.org/10.1016/j.dcan.2023.01.022">https://doi.org/10.1016/j.dcan.2023.01.022</ext-link></mixed-citation></ref>
</ref-list></back>
</article>