Gold serves as a key inflation hedge and portfolio stabilizer, making accurate price forecasting essential for investors. Hourly gold prices exhibit pronounced non-linearity and microstructure noise that limit traditional econometric models, motivating a shift toward deep learning. Existing CNN-LSTM architectures cascade convolutional and recurrent layers sequentially, without a mechanism to reconcile their complementary representations. We propose a Dual-Branch CNN-LSTM architecture with Gated Fusion, combining a convolutional-recurrent deep branch with a parallel raw-input skip branch, adaptively merged by a learned gate inspired by the Gated Multimodal Unit. Input sequences are restructured via sliding windows across three forecast horizons (24→1, 48→2, and 72→3 hours). The model is validated on 37,140 hourly XAUUSDm observations (Exness, 2020-2026) against 1D-CNN, LSTM, and Sequential CNN-LSTM baselines, achieving the best or tied-best accuracy across all configurations. For the 24-hour horizon, MAE = 9.00 USD, R² = 0.9994, and DA = 52.6%, on the original USD scale. Diebold-Mariano tests confirm significant gains over the 1D-CNN and Sequential CNN-LSTM baselines (p < 0.001), while McNemar tests confirm directional-accuracy gains in 8 of 9 comparisons (p < 0.05). These results show a modest but robust improvement over prior CNN-LSTM designs, offering a reliable foundation for risk management pending economic backtesting.
Accurate geolocation for low-power wide-area (LPWAN) devices is desirable when GNSS is unavailable or too energy-expensive, yet RSSI-/TDoA-based approaches are often fragile under channel variability, collisions and cross-device heterogeneity. We address this gap with a reproducible, tabular pipeline that maps LoRa RSSI/SNR/ToA and PHY metadata to 2D positions, compares strong tabular baselines (k-NN, Random Forest, LightGBM, XGBoost), and crucially evaluates them under group-aware (device-wise) splits to avoid identity leakage. On an ns-3-generated LoRa dataset of about 3.3 × 104 labeled receptions, Random Forest attains the tightest distribution with p50 ≈ 0 m and p95 < 1 m, whereas k-NN, despite a low median, exhibits a much heavier tail (p95 ≈ 187 m), underscoring the need to report both central and tail metrics. These results indicate that simple, edge-feasible models can perform gateway-side inference with robust accuracy when fed cleaned features and evaluated with realistic splits, making the approach attractive for practical LPWAN/IoT deployments.
INTRODUCTION: Elderly and disabled populations worldwide are growing faster than care systems can absorb them, creating strong demand for affordable home-automation tools that let residents remain safe at home without continuous caregiver presence. Most low-cost IoT prototypes reported in the literature cover only one or two functional subsystems and send alerts to a single recipient, leaving a practical gap for multi-subsystem, multi-recipient solutions. OBJECTIVES: To design, build, and test a modular home-automation system that brings together four specialised sensor nodes—radio-frequency access control, ultrasonic proximity detection, combustible-gas monitoring, and rain-triggered physical protection—plus a fifth node combining multi-recipient alert coordination with remote lighting control across five areas of the dwelling, all over a coordinated wireless network with real-time alerts sent to multiple users simultaneously. METHODS: A waterfall lifecycle combined with test-driven development guided component selection and firmware coding. Five ESP32 microcontrollers exchange data through a lightweight publish-subscribe protocol; each node handles its own control logic and actuator independently, while a companion mobile application allows remote supervision and manual override. RESULTS: All four subsystems were functionally validated through virtual simulation followed by physical demonstration on a scale dwelling model: the RFID module granted or denied access correctly for authorised and unauthorised cards, the ultrasonic module opened and closed the back door in response to detected presence, the gas sensor triggered visual, audible, and remote alerts above the configured threshold, and the rain sensor activated the laundry-protection servo on detected precipitation. The Telegram notification channel delivered alerts to the user with an observed average response time of under two seconds. Total component cost came to approximately 73 United States dollars, far below comparable commercial products. CONCLUSION: The five-node distributed architecture demonstrates functional feasibility and meets the cost target needed for practical use in urban homes in developing countries, and offers a reproducible testbed for future assistive-technology research that includes controlled reliability and latency testing.
The increasing prevalence of mosquito-borne diseases demands scalable, accurate, and energy-efficient surveillance systems capable of real-time operation in resource-constrained environments. This extended study presents an FPGA-optimized, drone-based mosquito breeding site detection framework leveraging quantized deep learning models for edge deployment. Building upon prior work, this paper introduces an enhanced implementation of YOLOv8-Tiny and YOLOv9-Small architectures, optimized through INT8 quantization, batch normalization folding, and FPGA-aware architectural refinements for execution on the PYNQ-Z2 platform. High-resolution aerial imagery acquired from unmanned aerial vehicles (UAVs) is processed in real time using the proposed system to identify and classify potential mosquito breeding sites such as stagnant water bodies and container habitats. Experimental evaluations carried out on six state-of-the-art object detection architectures showed that the proposed quantized YOLOv8-Tiny variant offers the optimal compromise among accuracy, speed, and energy efficiency by achieving 90.2% accuracy in the field, 20 FPS performance, and consuming 7.8 W of power. It is also proved that the proposed system reduces DSP and BRAM resources by more than 20% over the floating-point processor. The results obtained in the real-world deployment scenario proved the robustness of the system under different environmental situations. It is also found that the system can cover 10 km² of area per hour while providing a 44% reduction in operational costs over traditional methods. The proposed framework using FPGA acceleration and quantization awareness is useful for developing efficient vector surveillance systems.
Classical computing is approaching its physical limits, so quantum computing as an alternative model requires careful evaluation. This study compares various aspects of two systems by looking at formal algorithmic complexity, simulation-based comparisons with realistic noise models, quantitative meta-analysis, figuring out the difference between physical and logical error rates, and clearly defining contributions. A thorough study examined 24,100 papers published between 2015 and 2024 from leading scientific databases. A study of algorithmic complexity using Big-O notation confirmed that Shor's algorithm is faster than standard O(exp((64/9)²/³ (log N)²/³ (log log N)²/³)) by a super polynomial factor, Grover's algorithm is faster by a quadratic factor (O(√N) versus O(N)), and quantum simulation is faster by an exponential factor. There were Five quantum machines exhibited physical error rates ranging from 10⁻² to 10⁻⁵. Simulation-based testing revealed that the quantum advantage emerges above certain complexity thresholds: at extremely high complexity, quantum processing reached 10.3 μs versus 40.2 μs for classical computing, with better error resistance (2.3% for quantum vs. 5.9% for classical). Quantum computing has real benefits in terms of speed, mistake tolerance, and energy economy for problems that are too complicated to solve with traditional methods. However, significant challenges remain in scalability and workforce development. Hybrid quantum-classical models offer the most promising path to near-term practical benefits. On the other hand, the shift to postquantum security needs instant attention, regardless of when quantum hardware will be available.
Enhanced external counterpulsation (EECP) is a non-invasive cardiovascular rehabilitation therapy, but its follow-up in routine outpatient practice is often difficult to standardize, interpret efficiently, and communicate consistently. Although nailfold videocapillaroscopy (NVC) offers a direct view of peripheral microcirculation, its practical use remains constrained by image-quality variation, manual reading burden, and inconsistent reporting. To address these issues, this study adopts a Design Science Research perspective to develop a generative-AI–assisted medical information system for NVC-based EECP follow-up. The proposed artifact integrates image quality control, structured AI-assisted interpretation, report generation, human review, and audit logging into a governable outpatient workflow. A pilot implementation in a single clinic involving three physicians suggests that the system can reduce turnaround time and labor burden, support more scalable service delivery, and provide a structured basis for improving report consistency, though systematic consistency evaluation remains for future work. This study contributes an MIS artifact that demonstrates how generative AI can be operationally embedded into EECP follow-up.
In interconnected systems such as the Internet of Things (IoT), industrial control systems, and smart cities, real-time intrusion detection is crucial in smart network environments. It analyses network traffic in real time, enabling rapid detection and mitigation of cyber threats. With the help of artificial intelligence, including deep learning, real-time intrusion detection systems (IDS) can spot suspicious patterns, adapt to new-fangled attack vectors, and keep latency low, all without sacrificing system performance or reliability. This paper addresses intrusion detection in smart network environments by introducing Net Sentry DL, a deep learning framework. To extract spatio-temporal features and improve interpretability, the model uses a combination of CNNs, TCNs, and attention-guided fusion. Data imbalance and noise are addressed using an entropy-based pre-processing method and a class-preserving algorithm called Synthetic Minority Oversampling Technique (SMOTE). The UNSW-NB15, BoT-IoT, and TON_IoT benchmark datasets are used to assess Net Sentry DL. It surpasses models such as SVM, Random Forest, LSTM, GRU, and CNN in binary classification, reaching up to 0.99 accuracy and 0.98 F1-score on BoT-IoT. With TON_IoT, it achieves an accuracy of 0.95, and with BoT-IoT, up to 0.97, in multi-class configurations. Using SHAP, attention heatmaps, and gated fusion visualisations, the model exhibits strong explainability and robust generalisation, as demonstrated by cross-dataset testing. Through ONNX conversion and low quantisation loss (0.7%), it efficiently deploys and achieves low inference time (37ms/sample). The significance of each module, particularly CP-SMOTE and the TCN-attention combination, has been confirmed by ablation studies. For ever-changing IoT-based infrastructures, Net Sentry DL demonstrates competitive accuracy, interpretability, and deployment efficiency for intrusion detection.
INTRODUCTION: Blockchain technology has achieved widespread adoption across diverse application domains, yet its foundational consensus mechanisms remain largely static and may not suited to the dynamic, heterogeneous conditions of modern decentralized networks. While adaptive consensus has emerged as a promising solution, a comprehensive and systematic framework for classifying, evaluating, and selecting adaptive mechanisms remains notably absent.OBJECTIVES: This survey addresses this critical gap by introducing a structured taxonomy of adaptive consensus mechanisms for distributed blockchain systems, underpinned by a systematic review of performance evidence and real- world deployment experiences.METHODS: Following PRISMA 2020 guidelines, we conducted a systematic search of 1,675 records from Scopus (2020– 2025), ultimately including 68 peer-reviewed studies. We analyze four principal categories of adaptive consensus namely dynamic parameter adjustment, consensus algorithm switching, hybrid mechanisms, and AI/ML-enhanced protocols across five key performance dimensions: throughput, scalability, energy efficiency, security level, and implementation complexity. Case studies of SABEC for UAV coordination, 6G cognitive radio spectrum management, and supply chain networks illustrate real-world deployment trade-offs.RESULTS: Our comparative analysis reveals fundamental performance–security trade-offs inherent to each adaptive category. Dynamic parameter adjustment mechanisms offer low implementation complexity but limited scalability gains; algorithm-switching approaches achieve high throughput (100–1,000 TPS) at the cost of very high implementation complexity; hybrid schemes such as HyFlexChain demonstrate 112.5+ TPS under BFT mode with high security but remain at the prototype stage; and AI/ML-enhanced protocols show theoretical promise yet face critical security challenges including adversarial attacks and model poisoning that undermine system trustworthiness. The maturity assessment confirms that higher implementation complexity consistently correlates with limited real-world deployment.CONCLUSION: Our findings demonstrate that no universal adaptive consensus mechanism exists for blockchain applications. This structured, evidence-grounded survey provides an effective methodology for designing, evaluating, and selecting adaptive consensus solutions for diverse decentralized system deployments.
Urban traffic monitoring plays a crucial role in intelligent transportation systems. The development of surveillance camera networks has generated a large amount of image and video data that can be exploited for traffic flow detection, tracking, and analysis tasks. However, detecting and tracking vehicles from fixed traffic surveillance cameras still faces many challenges. The main challenges include obscured objects, small target size, and high traffic density. This study presents a deep learning-based traffic monitoring framework for detecting and tracking multiple objects in urban traffic monitoring systems. The proposed framework integrates YOLOv11 for vehicle detection and DeepSORT with a Kalman filter-based state estimation method for tracking multiple objects. In addition, the SAHI technique is integrated to investigate its ability to support the detection of small objects in traffic data. The research framework was evaluated using a dataset collected from traffic cameras in Thai Nguyen, Vietnam. Numerous test scenarios were conducted with varying traffic densities, observation distances, and camera viewing angles. Experimental results showed that the YOLOv11 configuration combined with DeepSORT achieved a processing speed of approximately 10.1 FPS; for object detection tasks, the model achieved an mAP@0.5 of 0.66. Simultaneously, experimental results show that the proposed framework can maintain vehicle detection and tracking across consecutive frames under varying observation conditions. In addition, the integration of SAHI techniques recorded an improvement in detecting small objects, with mAP@0.5 increasing from 0.66 to 0.70 and AP_S increasing from 0.29 to 0.40. The results obtained demonstrate the potential applicability of the proposed framework to traffic detection, tracking, and monitoring problems in urban environments.
This study shows a mixed quantum-classical guidance system that combines an eight-layer Quantum Approximate Optimization Algorithm (QAOA) with Proximal Policy Optimization (PPO) reinforcement learning for cities that do not have GNSS. The design changes the settings of the quantum circuit on the fly based on real-time data from the environment. This makes noise in cities less of a static mistake source and more of an optimization constraint. After testing on 47.3 kilometers of GNSS-denied paths in three cities using a three-tier DARPA QuANET-aligned protocol and NIST-traceable equipment, the results show that the average positional accuracy is 0.15 ± 0.03 meters, which is 70% better than high-grade GPS/INS systems. With adaptive correction, the framework keeps 98.2% of its coherence under ISO 16750-3 shaking and 95.3% of its operational reliability during 24-hour signal rejection. It also reduces drift by 82% compared to LiDAR-SLAM over a kilometer. It is statistically significant (F (3,1196) =87.3, p<0.001, ·²=0.38), and the effect sizes are big. The system passes SAE Level 4 power limits (45.2W), goes beyond DARPA QuANET 2025 goals, and gets Hybrid Readiness Level 7 approval. Each unit is expected to cost $210,000, and it creates the first policy-ready quantum navigation platform for smart city application.
Land registration and record management systems worldwide continue to face significant challenges, including document fraud, long processing times, and inefficient maintenance procedures. Traditional methods involve several technical limitations that reduce reliability and transparency. To address these issues, the proposed system leverages blockchain technology to improve process efficiency, data integrity, and security in land registration workflows. In the proposed framework, users upload property details and supporting land documents while initiating a sale. However, fraudulent document uploads remain a common issue, enabling sellers to receive payments using forged records without the buyer’s knowledge. To mitigate such risks, a Convolutional Neural Network (CNN) is integrated to authenticate and validate uploaded land documents before further processing. Only documents verified as authentic are stored on the blockchain. The government authority converts these validated documents into Non-Fungible Tokens (NFTs) and mints them on the Ethereum blockchain. A unique hash is generated for each document, enabling secure verification and traceability through platforms such as Etherscan. Once the documents are confirmed to be valid, the property is approved for sale, and the ownership transfer between the seller and buyer is securely executed through the blockchain enabled system. We evaluated the performance of proposed framework by considering both blockchain performance metrics and CNN evaluation metrics.
River Teesta is prone to weather fluctuations that leads to cloud burst or heavy rainfall resulting in overflowing of Teesta dam (27.0018057°N, 88.4404352°E) situated across the Teesta Basin region. Dam overflow is one of the major causes of sudden and uncontrolled release of water, which often leads to flooding in nearby downstream areas. Accurate flood mapping is important to reduce the damage caused by floods in low-lying areas. This study uses a U-Net deep learning model to identify flooded regions near Kalimpong in the Teesta River area using satellite images. The proposed method identifies flood-affected areas accurately and provides useful support for early warning and water management systems. The model was compared with existing methods such as Otsu thresholding, Random Forest, SegNet, and PSPNet using IoU, F1-score, precision, and recall values. The proposed U-Net model achieved 98% IoU, 98% F1-score, 99% precision, and 97% recall, showing better results than the other methods. The IoU value was 26% higher than Otsu thresholding, 18% higher than Random Forest, 10% higher than SegNet, and 8% higher than PSPNet. The training and validation graphs show stable learning with very little overfitting and steady performance during all 100 epochs. In addition, NDVI time-series analysis for Lachung city and the Lachung–Teesta meeting point showed important changes in vegetation that are useful for water resource management. These results show that the proposed U-Net model can identify flood areas effectively and can help improve early warning systems and safe dam management. However, the study is limited to Sentinel-2 image-based flood segmentation for the Teesta basin and does not include real-time hydrological inputs.
Wireless Sensor Networks (WSNs) have become a foundational technology across diverse domains, ranging from critical healthcare monitoring to large-scale environmental management. However, the severe energy constraints of sensor nodes remain a persistent bottleneck, threatening both operational efficiency and network longevity. While metaheuristic algorithms offer promising solutions, existing reviews often focus on isolated network layers or rely on outdated datasets. Addressing this gap, this Systematic Literature Review (SLR) analyzes 48 primary studies published between 2019 and 2024, offering a holistic taxonomy that integrates routing and clustering optimizations. The findings reveal that Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO) continue to dominate the field, each appearing in 23.5% of studies. However, a decisive shift is observed toward hybrid techniques such as Firefly–PSO and Grey Wolf Optimization variants which demonstrate enhanced adaptability in avoiding local optima, albeit at higher computational costs. Performance evaluations remain heavily simulation-driven, primarily focusing on energy consumption (31.2%), network lifetime (29.8%), and throughput (19.9%), while real-world validations in domains like Industrial IoT remain scarce. Furthermore, the review identifies emerging trends integrating Machine Learning, Edge Computing, and UAV-assisted routing into metaheuristic frameworks, signaling a transition toward more secure and multi-objective optimization strategies. This study concludes by highlighting critical open issues in fault tolerance, heterogeneous node management, and security-aware routing, providing a strategic roadmap for developing resilient, deployment-ready WSN solutions.
Increased deployment of IoT systems in industrial, healthcare, smart city, and home environments has expanded the attack surface and complexity of cyber threats. Though a plethora of detection techniques have emerged in literature in the last decade, an unforgivable absence of statistical rigors and compare-and-contrast analysis on the operational characteristics is apparent in the literature sets. This paper presents a statistical analytical review of various contemporary IoT threat detection methods across a wide array of architectures: classical machine learning, deep learning, federated learning, blockchain-based systems, quantum-enhanced frameworks, and hybrid models. The review employs a multidimensional evaluation strategy, extracting both qualitative and quantitative metrics from each study; these enable an objective comparison across heterogeneous systems, Standard performance parameters—Scalability, Delay, Time Complexity, Memory Complexity, Make span, and Analysis Efficiency—are tabulated; thus, an unfolding of a universal analytical framework with almost 300 data points exposes trade-offs, bottlenecks to efficiency, and constraints to deployment. Furthermore, evaluation of the application-specific techniques for healthcare, agriculture, and smart grids were conducted in relation to adaptability and domain specifications. The work identifies that hybrid deep networks (e.g., CNN-LSTM) provide better accuracy at higher computation cost, while TinyML and ensemble models present a trade-off factor for both detection accuracy versus hardware efficiency. In addition, whereas quantum and blockchain Integrated systems have shown to be solid in theory, they face practical impairments. Research gaps identified here lead the discussion on future directions, toward explainability, energy-aware design, and adversarial resilience, thus providing a tangible roadmap toward the next generation of secure IoT frameworks.
INTRODUCTION: AI development is driven by innovation and cultural contexts of collaboration. As AI and IoT systems shape global interaction, understanding cultural influences on technology perception is key for adaptive design and governance. This study compares personal values and AI acceptance among researchers in China and Germany – two leading yet culturally distinct ecosystems.OBJECTIVES: To identify value patterns supporting trustworthy AI and effective cross-cultural collaboration in research and IoT contexts.METHODS: A cross-national survey (n = 200) using the Portraits Value Questionnaire (PVQ) and the Digital Technology Acceptance Scale (DTAS) examined factors shaping AI perception.RESULTS: Chinese participants show higher AI acceptance and stress self-enhancement and conservation; Germans emphasize self-transcendence and greater caution.CONCLUSION: Findings inform culture-aware AI design, value-aligned governance, and intercultural collaboration.
INTRODUCTION: The Internet of Medical Things (IoMT) has expanded rapidly, with a growing number of medical devices becoming interconnected and increasingly integral to healthcare delivery. However, this expansion has also introduced significant cybersecurity risks, making IoMT networks vulnerable to sophisticated cyber-attacks that threaten patient data confidentiality and system reliability. OBJECTIVES: This study aims to develop a robust intrusion detection framework capable of accurately identifying both known and unknown cyber-attacks in IoMT environments while minimizing false positives and false negatives. METHODS: The proposed framework employs a Capsule Neural Network (CapsNet) to effectively capture spatial hierarchies and part–whole relationships in network traffic data. Additionally, the Theory of Association (TOA) is utilized for batch-size hyperparameter tuning to enhance learning efficiency and detection performance. The model is evaluated using standard performance metrics to assess its effectiveness in detecting malicious traffic. RESULTS: Experimental results demonstrate that the proposed Intrusion Detection System (IDS) achieves an accuracy of 98.37%, precision of 98.57%, recall of 98.17%, and an F1 score of 98.37%. These results indicate strong real-time detection capability with minimal false positives and false negatives. CONCLUSION: The findings highlight the effectiveness of integrating deep learning techniques, particularly CapsNet with TOA-based optimization, in strengthening cybersecurity for IoMT networks. The proposed IDS provides a secure and efficient solution for protecting healthcare data and ensuring patient confidentiality, offering a promising approach to enhancing the security and performance of healthcare IoMT systems.
INTRODUCTION: Adaptive Traffic Signal Optimisation (ATSO) is a challenging problem for urban traffic networks, having important implications for congestion reduction, traffic efficiency, and environmental conservation. Conventional traffic signal control techniques, i.e., fixed-time and rule-based control, fail to respond to dynamic traffic behaviour efficiently. OBJECTIVES: Recent developments in Reinforcement Learning (RL) have been promising for ATSO but are plagued by poor scalability, lack of coordination in multi-intersection networks, and inefficiency in dealing with continuous action spaces. METHODS: Furthermore, most RL-based solutions are based on simplistic state representation and fail to incorporate complex interdependencies between traffic signals. Considering these limitations, this paper introduces a new framework, Multi-Agent Soft Actor-Critic with Graph Attention Networks (MASAC-GAT), which unites the sample efficiency and stability of Soft Actor-Critic (SAC) with the relational modelling ability of Graph Attention Networks (GATs). RESULTS: The proposed method exhibited significant performance gains on three important traffic metrics: Signal Adjustment Efficiency (92%), Average Waiting Time (20–35 seconds), and Congestion Prediction Accuracy (93%), outperforming DQL, PPO, A2C, GNN-based variants, and knowledge sharing DDPG (KS-DDPG). Through minimised redundant signal changes and reduced vehicle delays, the method ushers in the next generation of smart transportation systems. CONCLUSION: The proposed method facilitates decentralised yet coordinated control of traffic signals by utilising local observations and global context. The proposed method unites real-time traffic observations, e.g., traffic volume, vehicle speeds, weather, accident reports, and signal status, into a customised OpenAI Gym environment for training and evaluation.
Modern Internet of Things (IoT) and communication networks operate under dynamic, large-scale, and resource-constrained conditions where conventional control can falter. This survey synthesizes 108 peer-reviewed (including preprints) studies (2007–2025) on nature-inspired control mapped to the TCP/IP stack across the classes of network functions. The survey introduces a three-axis taxonomy (TCP/IP layer × biological metaphor × function) and a unified KPI scheme with a normalization formula to compare heterogeneous reports. Aggregated evidence indicates typical gains of 20–50% energy reduction, 25–40% delivery-ratio improvement, 10–35% latency reduction, and > 0.95 F1 for immune-inspired intrusion detection, when compared to canonical baselines. This survey further makes explicit how few studies validate on real hardware or under adversarial conditions. Moreover, this survey analyzes real-time constraints, hyperparameter sensitivity, and integration pathways with 6LoWPAN/RPL, TSCH/6TiSCH, MQTT, and CoAP, outlining steps toward deployable, explainable, and deterministic nature-inspired control.
The rapid expansion of the Internet of Things (IoT) has intensified the challenge of achieving dynamic bandwidth allocation while maintaining security across heterogeneous devices and communication protocols. Conventional static allocation schemes lack adaptability, while existing learning-based or blockchain-based approaches typically optimize performance or trust in isolation. To address this gap, this paper proposes a hybrid framework that integrates Q-learning–based adaptive bandwidth allocation with a lightweight, permissioned blockchain-based trust mechanism. The framework is evaluated through MATLAB-based simulations involving 100 heterogeneous IoT devices under dynamic traffic conditions and adversarial behavior. Performance is compared against multiple baselines, including static allocation, learning-only and blockchain-only schemes, classical scheduling algorithms (WFQ and DRR), and a deep reinforcement learning approach (DQN). The results reveal clear trade-offs among bandwidth utilization, fairness, energy consumption, and security. Static and classical schedulers provide predictable fairness but remain vulnerable to malicious activity. Learning-only and deep reinforcement learning approaches improve adaptability but lack intrinsic trust awareness, while blockchain-only enforcement enhances security at the expense of responsiveness. By coupling adaptive decision-making with trust validation, the proposed hybrid framework achieves a balanced operating point, offering stable bandwidth utilization, improved energy efficiency, and robust attack resilience under noisy and uncertain conditions. These findings highlight the importance of aligning learning mechanisms with trust-aware constraints for secure and scalable bandwidth management in heterogeneous IoT networks.
5G networks are complex. They must handle different types of connections. These networks support industries, cities and mobile users. Managing traffic is difficult. Traditional methods are not efficient. Open RAN (O-RAN) is a new approach. It allows better control of network functions. It helps improve user experience. This is possible through automation and artificial intelligence (AI). AI helps make smart decisions in real time. This paper introduces a software framework called ns-O-RAN. It combines a real-world RAN Intelligent Controller with a network simulator. This allows testing AI solutions without expensive hardware. The study also proposes a smart handover method. Handover is the process of switching users between base stations. The goal is to reduce delays and improve speed. The new method uses deep reinforcement learning (DRL). DRL learns the best way to assign users to base stations. The framework collects a large amount of data. It trains the AI system using this data. The model learns from past network conditions. It then makes better decisions for the future. The proposed solution increases network efficiency. The researchers tested their model. They compared it with traditional handover methods. This means faster speeds and fewer connection losses. The framework also enables real-time monitoring. It detects network issues quickly and adapts to changing conditions. This ensures users get stable and high-quality connections. Additionally, this approach supports different types of applications. It works well for video streaming, voice calls and industrial automation. This work has important implications. It helps telecom providers improve service quality. It also reduces operational costs. Researchers and engineers can use this framework for further development. This study contributes to the future of AI-driven mobile networks.
