Volume 17, Issue 3, July 2026

Unified Edge–Cloud–Agentic Taxonomy for LLM-Enabled IoT Systems: A Systematic Critical Review of Intelligent Edge Ecosystems

Pages: 1-10 (10) | [Full Text] PDF (723K)

Rana Zubair Ahmad, Muhammad Junaid Arshad

International Institute of Science, Arts and Technology (IISAT) University of Engineering and Technology (UET), Lahore

Abstract - Abstract—In the last couple of years, there has been a change in the way intelligent IoT systems are being considered— and built. The large language models appeared. Autonomous agents followed. Edge hardware improved, models got smaller, and then deployments that previously were forced to go through the cloud began to appear. like they might be able to make real local reasoning. The field moved fast. What did not keep pace was any coherent framework for making sense of it. We conducted this review because that gap was frustrating us in our own work. Every time we tried to compare two intelligent IoT architectures—one federated, one agentic, say, or one TinyLLM deployment against a RAG-augmented edge system—we found the existing vocabulary inadequate. The dimensions that seemed to matter most in practice (where reasoning actually happens, how autonomous the system really is, whether it can be trusted in a safety-critical setting) were precisely the ones that existing surveys glossed over or ignored entirely. What this paper offers is a systematic critical review across eight thematic areas, and from that review, a ninedimensional taxonomy we call the Unified Edge–Cloud–Agentic Taxonomy. The nine dimensions cover intelligence placement, edge capability level, cloud reliance, reasoning autonomy, orchestration design, federated collaboration, agentic coordination, sustainability orientation, and trustworthiness. Together, they give researchers and practitioners a shared language for the architecture space that exists today—not the simpler one that existed five years ago.

Index Terms - Internet of Things, Large Language Models, Edge Intelligence, TinyLLMs, Federated Intelligence, RAG, Agentic AI, Multi-Agent Systems, Edge–Cloud Computing and Taxonomy

SafeSpeak: A Real-Time Message Filtering System for Safe Guarding Against Toxicity using BERT

Pages: 11-16 (6) | [Full Text] PDF (423K)

Renuka Garad, Riddhi Lohiya, Shradha Jamge

Dept. of Information Technology GHRCEM, Pune, India

Abstract - A huge number of user-generated messages are created each day on social media platforms. Messages may con- tain abusive or toxic messages which could cause cyberbullying, mental harassment, and emotional disturbance. The keyword- based approach fails to understand the context and is incapable of handling different languages such as english, hindi, marathi. This paper showcases SafeSpeak, which is a real-time message filtering system based on a Bidirectional Encoder Representations from Transformers (BERT) based transformer approach which would prevent toxic messages from being posted. SafeSpeak handles multiple languages as well as mixed languages such as english, hindi, and marathi, and tests messages at the time of submission to prevent the posting of abusive messages. We are using BERT to create our system. Using BERT gives us better results because the system is able to comprehend the context as well as be more accurate. The system has a good accuracy rate as well as an adequate response time which will make it effective for real-life application.

Index Terms - Content Moderation, Multilingual NLP, Trans- Former, BERT, Sequence Classification, Social Media, Toxicity Detection

Ensuring Service Reliability in Cloud Platforms: A Systematic Review of QoS Metrics and SLA Strategies

Pages: 17-26 (10) | [Full Text] PDF (588K)

Aiman Farooq

Department of Computer Science, University of Engineering and Technology

Abstract - The convergence of cloud computing and the Internet of Things (IoT) has created an unprecedented demand for reliable, high-performance service delivery across distributed computing environments. As billions of IoT devices generate massive volumes of data requiring real-time processing, storage, and analytics, ensuring consistent Quality of Service (QoS) and enforcing Service Level Agreements (SLAs) have become critical challenges. This paper presents a comprehensive systematic review of research conducted between 2010 and 2025 on QoS metrics and SLA management strategies in cloud computing platforms. Through an extensive analysis of over 100 studies, we classify existing approaches into key categories including QoS modeling techniques, SLA lifecycle management, resource scheduling algorithms, machine learning-based prediction models, and blockchain-enabled SLA enforcement. Comparative tables summarize the methodologies, evaluation metrics, tools, and contributions of the reviewed works. The paper identifies persistent research gaps in areas such as real-time SLA monitoring for edge-fog-cloud continuum, standardization of QoS metrics across heterogeneous platforms, and scalability of intelligent resource management frameworks. This review serves as a consolidated reference for researchers and practitioners seeking to understand the evolution, current state, and future directions of QoS and SLA research in cloud-IoT ecosystems.

Index Terms - Cloud Computing, Quality of Service, Service Level Agreement, IoT, Fog Computing, Edge Computing, Resource Management, Machine Learning and Systematic Review

Development of an IoT-Based Smart Energy Management Controller for Dynamic Electricity Pricing

Pages: 27-32 (6) | [Full Text] PDF (389K)

Najaf Al Hadi, Massod Hasan, Aziz Wahbi

Computer Science Department, Sudan University, Sudan

Abstract - This paper proposes an IoT-enabled Smart Energy Controller for Residential Air Conditioners to reduce both electricity consumption and electricity costs through a novel Demand Response-based control algorithm. A hardware prototype of the Smart Energy Controller, along with its supporting software, has been developed to demonstrate the proposed Energy Management System for intelligent thermostat control of residential air conditioners. The proposed Demand Response algorithm is evaluated under three different operating strategies. The results demonstrate that the Real-Time Pricing-based strategy achieves the highest reduction in both electricity consumption and operating cost, outperforming the other proposed approaches.

Index Terms - IoT-Based Smart Energy System, Efficiency, Electricity Pricing, Strategy and Outcomes

A Progressive Optimization Technique for MobileNetV3 in Rice Plant Disease Detection

Pages: 33-44 (12) | [Full Text] PDF (690K)

Haseeb Hassan Ansasri, M. Junaid Arshad

Department of Computer Science, University of Engineering and Technology (UET), Lahore, Pakistan

Abstract - The identification of disease problems in paddy is very crucial for achieving good food security levels and good crop yield rates. The traditional methods used for manual recognition of such diseases have been identified as being highly inefficient and subject to human errors. This paper suggests an improved deep learning model for the automation of the detection of different paddy plant diseases such as Rice Blast. We have tested our hypothesis using 1440 different image data and have tried out some of the latest deep learning models available such as DenseNet121, Xception, and EfficientNetB4. However, we have focused on one particular model, namely Optimized MobileNetV3 Large that managed to achieve an accuracy rate of 96.85%. Additionally, the model was capable of running very efficiently and was found to consume only 523 MB RAM.

Index Terms - Rice Plant Disease, Deep Learning, Optimized Network, MobileNetV3 and Accuracy

A Secure Federated Learning-Based Framework for Real-Time Cyber Threat Detection in 5G-Enabled IoT Environments

Pages: 45-54 (10) | [Full Text] PDF (528K)

Awais Rasool, Zartasha Kiran, Muhammad Junaid Arshad

Department of Software Engineering, The University of Lahore, Pakistan Department of Computer Science, University of Engineering and Technology Lahore, Pakistan

Abstract - The spread of 5G-enabled Internet of Things (IoT) ecosystems has enhanced connectivity, scalability and real-time communication in smart environments to a great extent. Yet, they have also ushered in some new cybersecurity threats because of the distributed nature of IoT devices, high data velocity and added attack surface. Traditional centralized intrusion detection systems are inadequate to perform real-time anomaly detection, because of latency and privacy concerns and scalability. To solve these problems in this study a federated learning-based framework is proposed for real-time anomaly detection in 5G supported IoT environments, named Edge-FL Guard. The framework combines the use of lightweight deep learning models such as autoencoders and LSTM networks at the edge nodes for local inference with the collaborative training of models using federated learning without data sharing. This way the preservation of privacy would not only minimize the overhead of communication but also promote scalability among the heterogeneous devices of the IoT. The effectiveness of the proposed framework was evaluated with the CICIDS2017 dataset, TON_IoT and synthetic datasets under different cyberattack scenarios. A full comparative study of the best practices in the literature shows that hybrid deep learning and federated learning algorithms yield better results. Among others, the hybrid model presented the best performance in terms of accuracy 97.12% and ultra-low latency 0.0476 ms outperforming the others. The results show that the federated learning scheme with hybrid deep learning models is an effective approach to boost the accuracy of the anomaly detection in real time. The results verify the efficiency, security and scalability of the Edge-FL Guard solution and that hybrid FL-based architectures are the best way forward for next-generation IoT cybersecurity solutions.

Index Terms - Federated Learning, Edge AI, IoT Security, 5G Networks, Anomaly Detection, Privacy-Preserving and Machine Learning

Intelligent IoT Traffic Engineering and Monitoring: A Review of AI, SDN, and QoS Approaches

Pages: 55-64 (10) | [Full Text] PDF (488K)

Hassan Ahmad

Department of Computer Science, University of Engineering and Technology Lahore, Pakistan

Abstract - The Internet of Things (IoT) has changed the way devices communicate, generating huge amounts of network traffic from billions of connected devices. Managing this traffic — through engineering, metering, and monitoring — is critical to ensure network performance, security, and quality of service (QoS). Over the past decade (2015–2025), researchers have explored many techniques including machine learning (ML), deep learning (DL), Software-Defined Networking (SDN), edge/fog computing, and federated learning to tackle these challenges. This review paper provides a complete overview of the research done in the domain of IoT traffic engineering, metering, and monitoring. We organize the existing work into clear categories, present comparison tables, discuss popular datasets, highlight key trends, and point out open challenges. The goal of this paper is to serve as a one-stop reference for anyone who wants to understand what has been achieved in this field over the last ten years and what directions future research should take.

Index Terms - Internet of Things, Traffic Engineering, Traffic Monitoring, Traffic Classification, Software-Defined Networking, Deep Learning, Machine Learning, Quality of Service, Anomaly Detection and Federated Learning

IoT-Driven Predictive Maintenance in Industrial Applications: Technologies, Frameworks, and Future Challenges

Pages: 65-72 (8) | [Full Text] PDF (423K)

Muhammad Sarmad Javed

Institute of Data Science, University of Engineering and Technology, Lahore

Abstract - Industries are increasingly substituting reactive and schedule-based maintenance with predictive maintenance (PdM). Rather than waiting for breakdowns, today’s systems use sensors, IoT connectivity and artificial intelligence to continuously monitor the health of equipment and identify problems before they occur. This change reduces downtime and improves reliability and reduces maintenance costs. This paper reviews recent works on IoT-driven predictive maintenance for induction motors, rotating machinery, industrial automation systems, smart grids and other industrial assets. It explores the integration of sensors, wireless communication, cloud platforms, edge computing, digital twins and machine-learning models for smarter maintenance systems.The review discusses popular machine-learning approaches such as XGBoost, Random Forest, Support Vector Machines (SVM) and Convolutional Neural Networks (CNN). Several studies reported diagnostic accuracy >90%, especially for bearing and motor fault detection. The paper also presents a unified five-layer architecture for IIoT-based predictive maintenance systems.In addition to current advances, the review points out some ongoing challenges, such as model interpretability, cybersecurity risks, lack of public industrial datasets, and difficulty of deploying AI models directly on low-power edge devices. The paper concludes with the future research directions on explainable AI, TinyML, digital twins and secure industrial IoT systems.

Index Terms - Industrial Internet of Things, Predictive Maintenance, Induction Motors, Digital Twin, Machine Learning, Fault Diagnosis, Rotating Machinery and Industry 4.0.

A Comprehensive Study of Smart Wearable Devices for Physical fitness in IoT

Pages: 73-82 (10) | [Full Text] PDF (613K)

Syed Khuram Abbas, Hussain Mehdi

Department of Computer Science, University of Engineering and Technology (UET), Lahore, Pakistan

Abstract - In this paper, we describe wearable technology. Wearable technology has revolutionized health and fitness monitoring by providing real-time tracking of physiological parameters. This paper reviews various wearable systems, including smart watches (SM), fitness trackers, smart clothing and biosensors, highlighting their applications in health monitoring, physical activity tracking, chronic disease management, and rehabilitation. It discusses the technological advancements, challenges, and future prospects of wearable health and fitness systems. Smart devices, such as smart watches and fitness bands, are widely used to monitor cardiovascular endurance, muscular strength, muscular endurance, flexibility, and body composition, leveraging sensors for heart rate, motion, and bio impedance, alongside AI-driven insights. These devices offer accessible fitness tracking but face significant limitations that hinder inclusivity across diverse populations. Despite their popularity, technical limitations undermine their inclusivity for diverse populations. Optical heart rate and bio impedance sensors falter with darker skin tones, tattoos, or hormonal fluctuations, yielding inaccurate data for older adults (56+), females, and non-binary/transgender users on hormone therapy. Motion sensors misinterpret movements for disabled users with tremors or limb differences, skewing strength and endurance metrics. The review also explores issues related to data accuracy, privacy, and user acceptance.

Index Terms - Wearable Systems, Health Monitoring, Body Composition, Muscular Strength, Fitness Tracking, Biosensors and Smart Watches

Smart Meter Security: An Analysis of AMI Vulnerabilities and Defense Mechanisms

Pages: 83-92 (10) | [Full Text] PDF (416K)

Sikandar Ahmad

Computer Science Department, University of Engineering & Technology, Lahore, Pakistan

Abstract - The structure of the world energy is radically changing to the digitally interrelated Smart Grid in place of a unidirectional legacy grid system. The turning point of this development is the Advanced Metering Infrastructure (AMI) that is a very complicated consortium of both smart meters and communications networks that can initiate a real-time and two-way data desk activity between utility providers and consumers. That notwithstanding, physical electrical technology and open cyberspace networks do pose a security dilemma where the open network is exposed to a lot of cyber-attacks through their mere exposure to the creation of efficiency. In this review paper, a detailed analysis of the cybersecurity environment that accompanies AMI will be provided, and it isolates architectural vulnerabilities that predispose smart meters to the cyber warfare. We much consider the threat vectors that have high impact, that is, we discuss the debilitating effect of network availability of Denial of Service (DoS) attacks and data integrity of False Data Injection (FDI) moves, to plunder energy or load balancing. Further, the paper also discusses the fact that Non-Intrusive Load Monitoring (NILM) is a critical infringement of consumer privacy which is designed to transform small data points of consumption into weaponized assets that could be deployed to develop intimate user observations. In response to these shortcomings, we put the levels of effectiveness of the new defense strategies against traditional cryptographical paradigms and the next generation application of our new encryption schemes, incorporating Machine Learning-based Intrusion Detection Systems (IDS) and decentralized authentication models based on Blockchain. In conclusion, this review concludes that the implementation of the Smart grid should not be reduced to traditional perimeter-based defenses but rather dynamic and comprehensive approach of defensive hardening of physical computing device, highly advanced encryption and the capacity to identify abnormalities automatically as randomised to resilient to important energy infrastructure.

Index Terms - Security Analysis, Defense Mechanisms, Advanced Metering System, Cyberspace and Energy Infrastructure