An intelligent technique based on jellyfish algorithm for priority-based task scheduling in IoT/Fog networks

per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-05241113articleAn intelligent technique based on jellyfish algorithm for priority-based task scheduling in IoT/Fog networksS. Sohrabissohrabi69@yahoo.com1M. Sakhaeisakhaei@basu.ac.ir2M. Nassirim.nassiri@basu.ac.ir3R. Mohammadir.mohammadi@basu.ac.ir4Dept. of Comp. Eng., Faculty of Engineering, Bu-Ali Sina University, Hamedan, IranDept. of Comp. Eng., Faculty of Engineering, Bu-Ali Sina University, Hamedan, IranDept. of Comp. Eng., Faculty of Engineering, Bu-Ali Sina University, Hamedan, IranDept. of Comp. Eng., Faculty of Engineering, Bu-Ali Sina University, Hamedan, IranFog computing was implemented to enhance the service quality of IoT applications in cloud computing. The increase in latency and bandwidth overhead in cloud computing is a result of IoT devices transmitting large amounts of data. IoT devices are responsible for producing significant volumes of data that need to be managed, resulting in higher demands on the fog-cloud computing network and necessitating resource management in this setting. In the fog-cloud computing system, tasks of varying priorities can be dispatched for processing. Providing an appropriate scheduling strategy for tasks in the fog layer is challenging due to the multitude of tasks and resources involved. Task scheduling in fog-cloud computing systems aims to optimize task allocation and execution while considering relevant constraints. Thus, this research suggests an enhanced method using the jellyfish algorithm and genetic algorithm to minimize energy consumption and execution time while considering task priority, which is crucial in task allocation. The results of the simulation and comparison of the suggested algorithm with other algorithms in this study support the effectiveness of the proposed method in decreasing both execution time and energy usage.http://ijece.org/Article/48922Internet of Thing Fog Computing Resource Allocation Algorithm Simulated Annealing Algorithm Jellyfish Algorithmper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-052411426articleA Resource Management Method for Fog-DSDN Networks Using Microservices Architecture and Echo State Networks (ESN)A. Payamaniabbas.payamani1366505@gmail.com1Ehsan Matinfarematinfar@gmail.com2E. Maghamie.maghami@el.iut.ac.ir3M. JanbozorgiMohammad.janbozorgi@yahoo.com4Elec. and Comp. Faculty, Dorud Branch, National University of Skill, Dorud, IranDept. of Comp., Dorud Branch, Islamic Azad University, Dorud, IranSchool of Electrical and Computer Engineering,Elec. and Comp. Faculty, Dorud Branch, National University of Skill, Dorud, IranIn Fog–DSDN (Software Defined Distributed Network) environments, optimal resource allocation among multiple nodes with temporal, processing, and memory constraints represents a fundamental challenge. Existing approaches often lack adaptive mechanisms to address rapidly changing network conditions, resulting in increased service latency and reduced resource efficiency. This research presents an innovative method based on a microservices architecture and an Echo State Network (ESN) for managing and optimizing resource allocation in Fog–DSDN. In this approach, each microservice independently handles local data collection and processing, with its output aggregated as an Information Map at the fog controller level. Subsequently, the ESN model learns temporal traffic patterns to predict future node processing loads, enabling adaptive resource allocation decisions. A dual-layer design of input and output queues within each information cell further reduces processing congestion and improves system response time. To evaluate the performance of the proposed method, simulations were conducted in OMNeT++ across diverse traffic scenarios, and results were compared against two baseline systems: a microservices architecture without load prediction, and a base non-hierarchical Fog-DSDN model employing the TFS resource management method. Experimental results demonstrate that the proposed method achieves, on average, a 12.57% improvement in processing resource efficiency, an 18.85% improvement in memory utilization, and a 13.39% reduction in service latency compared to the two baseline architectures and the TFS resource management approach. These findings indicate that the combination of a modular microservices structure with intelligent load prediction using ESN can provide an efficient, scalable, and lightweight solution for resource management in Fog–DSDN.http://ijece.org/Article/49971 Fog computing software-defined distributed networks (SDDN) resource management deep learning.per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-052412737articleDesign a Novel Emotion Assessment Approach for Cancer Care Based on Large Language ModelsN. Fareghzadehfareghzadeh@iau.ac.ir1M. Ghobadim.ghobadi@iauec.ac.ir2P. Rahmanirahmani.engineer@gmail.com3M. Bazarganimbzirn@iau.ac.ir4Dept. of Comp. Eng., Khodabandeh Branch, Islamic Azad University, Khodabandeh, IranFaculty of Computer Science, Electronics Department, Islamic Azad University, Tehran, IranFaculty of Computer Science, Pardis Branch, Islamic Azad University, Pardis, IranDept. of Comp. Eng., Zanjan Branch, Islamic Azad University, Zanjan, Iran  <div class="lRu31" dir="ltr">This research initiated to address the concern and challenge of the lack of specialized native tools in the field of emotion analysis in cancer patients and to provide a new system for assessing the emotions of these patients. The main goal of this study was to design and implement a hybrid system based on deep learning to identify the presence of one or more emotions simultaneously from six emotional categories (happiness, sadness, anger, fear, hope, despair) in Persian texts on social networks. This new approach allows for a better understanding of the complexity and simultaneity of emotions. For this purpose, a set of 10,000 cancer-related posts from social platforms was tagged. The proposed model, which uses the intelligent integration of the large native language model ParsBERT with the Bi-GRU network, was fine-tuned for this task. The evaluation results clearly demonstrate the remarkable achievement of this research; the proposed method achieved an accuracy of 86.8% and an average F1-Score of 84%. The proposed system showed an improvement of 2.3% in accuracy compared to the baseline models. The system performs effectively in detecting frequent emotions such as sadness with an F1-Score of 92% and despair with an F1-Score of 90%, providing an automated and powerful tool for monitoring the emotions of cancer patients and more effective psychotherapy interventions. <div class="OvtS8d"> </div> <div class="NQSJo"> </div> </div> <div class="UdTY9 WdefRb" data-location="2"> </div>http://ijece.org/Article/51872Natural Language Processing Emotion Analysis Cancer Large Language Models Deep Learningper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-052413848articleLightweight Hybrid Framework for IoT Security Using Optimized Random Forest and Adaptive Feature Selection in Edge-Cloud ArchitectureMohsen Ashrafim-ashrafi@tvu.ac.ir1Security in Internet of Things (IoT) infrastructures has become a critical challenge due to the rising cyber threats, such as Denial-of-Service (DoS) attacks, data breaches, and device manipulation. IoT devices, as core components of these systems, are highly vulnerable to various attacks and anomalies due to their continuous network connectivity and handling of sensitive data, potentially leading to system disruptions, confidential data leaks, and financial or human losses. Despite significant advancements in IoT security, challenges such as the diversity and complexity of attacks, heterogeneous and voluminous sensor data, variations in normal system behavior, scarcity of high-quality training data, and the need for scalable methods continue to hinder accurate and timely attack detection. This study proposes a lightweight and intelligent hybrid framework based on an Optimized Random Forest (ORF) and adaptive feature selection to enhance IoT system security. The performance of various machine learning algorithms, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Artificial Neural Network (ANN), and ORF, was evaluated using the DS2OS dataset. Results demonstrated that all models achieved accuracies ranging from 0.9844 to 0.9948, with RF and ORF exhibiting superior performance at an accuracy of 0.9943, precision of 0.9943, recall of 0.9943, and an F1-score of 0.9937. Furthermore, integrating the Particle Swarm Optimization (PSO) algorithm reduced the false positive rate to 0.57%, while the Edge-Cloud architecture improved processing time by 40%. Compared to existing approaches, the proposed method reduced memory consumption by 29%, offering a scalable solution for IoT security.http://ijece.org/Article/51262Internet of Things machine learning anomaly detection security optimized random forest (ORF) edge-cloud architecture.per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-052414957articleA Multi-Objective Metaheuristic Approach for Improving Coverage and Connectivity in Wireless Sensor NetworksM. Basirnezhadbasir80@gmail.com1M. Houshmandmahboobehhoushmand@yahoo.com2S. A. Hosseiniabed_hosseyni@yahoo.com3M. Jalalimehrdad.jalali@srh.de4گروه مهندسی کامپیوتر، واحد مشهد، دانشگاه آزاد اسلامی، مشهد، ایران1Department of Electrical Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran.گروه علم‌داده کاربردی و هوش‌مصنوعی، دانشگاه SRH هایدلبرگ، هایدلبرگ، آلمانThis paper investigates and presents a new optimization algorithm titled Multi-Objective Cheetah Optimizer (MOCO), developed with the aim of increasing coverage and improving connectivity in wireless sensor networks. Utilizing metaheuristic optimization concepts, this algorithm offers a novel approach for optimally determining sensor node positions and managing resources. In this method, by defining a multi-criteria objective function including maximizing area coverage, improving connectivity between nodes, and reducing energy consumption, it is able to establish a suitable balance among these objectives. The proposed algorithm is inspired by the random search capabilities and the acceleration of the cheetah algorithm, enabling more effective searching in the solution space. The results of simulations and performed experiments show that MOCO, compared to three algorithms CSSO, MOFAC-GA-PSO, and HHA, has been able to increase the environment coverage rate to over 94.5%, raise the network connectivity rate to approximately 97.8%, and simultaneously reduce the average energy consumption of sensors by up to 15% compared to other algorithms.http://ijece.org/Article/52048Connectivity metaheuristic algorithms coverage wireless sensor networksper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-052415866articleA Semi-Supervised Learning Framework for Accurate Test Case Classification Using Language Embeddings and Semantic Text FeaturesMohammadHoseein Parvanehm.hoparvaneh@gmail.com1maryam nooraei S.nooraei@gmail.com2Department of Computer Engineering, Arvand International Branch, Islamic Azad University, Abadan, IranWith the growing importance of integrating artificial intelligence and software testing, moving toward the intelligent automation of evaluation processes and exam item classification has become an essential necessity. One of the key challenges in this domain is the strong dependency on labeled data, the production of which is costly and time-consuming. In this study, a semi-supervised learning framework was designed and implemented using pseudo-labeling to incorporate unlabeled data into the training process and weighting the unsupervised loss. The dataset used was AG News, consisting of four news categories, where only 20% of the data was considered labeled and 80% unlabeled. For feature extraction, the BERT-base model was employed as a language embedder, producing 768-dimensional vectors (default configuration). Data preprocessing included tokenization with BertTokenizer, removal of punctuation and irrelevant characters, and text normalization. Performance evaluation using Accuracy, Precision, Recall, and F1-Score demonstrated that the semi-supervised approach outperformed the supervised SVM under limited labeled data conditions, achieving an average improvement of 5–10% across the metrics.http://ijece.org/Article/51312Semi-supervised learning NLP semantic learning S3VM SVM