per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234219232articlemohsen Eghbalimohsen.eghbali@gmail.com1M. Mollakhalili Meybodimeybodi@gmail.com2kamal mirzaiek.mirzaie@maybodiau.ac.ir3Dept. of Comp. Eng., Maybod Branch, Islamic Azad University, Maybod, IranDept. of Comp. Eng., Maybod Branch, Islamic Azad University, Maybod, IranDept. of Comp. Eng., Maybod Branch, Islamic Azad University, Maybod, Iran<p style="direction: ltr;">Today, with the increase in mobile devices, malware has also spread to the Android platform. These malware are written in more complex ways that are difficult to detect. Machine learning and deep learning methods are used to detect them because they can identify complex malware patterns. One challenge in malware detection with machine learning and deep learning methods is the high dimensionality of training samples. In this paper, a binary version of the Black-winged kite (BKA) algorithm is presented to reduce the dimensionality of training samples for the detection of Android malware. In the proposed method, the first stage extracts malware features using the BKA algorithm, which are then fed to the LSTM neural network. The LSTM's role is to classify Android malware samples as benign or malignant. To improve LSTM accuracy, its meta-parameters are also optimized using an Arithmetic optimization algorithm (AOA). Experiments on the CICandMal2017 dataset showed that the proposed method achieved accuracies of 98.63%, 98.29%, and 97.48% for accuracy, sensitivity, and precision, respectively. In the proposed approach, when balancing is performed using the GAN method on the CICandMal2017 dataset, the average accuracy, sensitivity, and precision of the proposed method increase to 99.62%, 98.93%, and 98.52%, respectively. Experiments show that the proposed method is more accurate at detecting malware than dimensionality-reduction methods such as WOA, HHO, and AVOA. The proposed method is about 16.4% more accurate than the LSTM neural network.</p>http://ijece.org/Article/46626Machine learning deep learning feature selection dimensionality reduction black-winged kite (BKA) algorithm Android malware.per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234233245articleMa. Asadolahzade Kermanshahim_asadolahzade@comp.iust.ac.ir1A. Akbari Aziraniakbari@iust.ac.ir2B. Nasersharifbnasersharif@kntu.ac.ir3School of Computer Engineering, Iran University of Science and Technology, Tehran, IranSchool of Computer Engineering, Iran University of Science and Technology, Tehran, IranDepartment of Computer Engineering, K. N. Toosi University of Technology, Tehran, Iran<p style="direction: ltr;">Data scarcity is the main challenge for DNN-based speech recognition, and data augmentation serves as an effective solution. This paper presents a comprehensive taxonomy of data augmentation methods in speech recognition while investigating the effectiveness of the most important techniques in this domain, masking-based methods, under limited data conditions. The examined methods include two powerful approaches: SpecAugment and word masking. Despite their proven effectiveness in high-resource scenarios, these methods have been less studied under limited data conditions. After analyzing the shortcomings of word masking in limited data settings, we propose two novel methods: (1) Gradual masking, which begins training with frame-level masking and then transitions to word-level masking; and (2) Word frequency-aware masking, which masks high-frequency words first, followed by low-frequency words. Experiments on the 100-hour LibriSpeech subset demonstrate that our first proposed method achieves a WER of 6.8% on the clean test set and 18.2% on the challenging test set, representing improvements of 6.8% and 4.2% respectively over SpecAugment. The second proposed method reaches a WER of 6.6% on the clean test set and 17.3% on the challenging test set, achieving improvements of 9.6% and 8.9% respectively compared to SpecAugment.</p>http://ijece.org/Article/50484Speech recognition word masking data augmentation limited data.per Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234246256articleZ. Raisizobeir.raisi@cmu.ac.ir1R. Damanidamani@cmu.ac.ir2E. Saranisarani@cmu.ac.ir3V. Nazarhzehi Hadv.nazar@cmu.ac.ir4Elec. Eng. Dept., Marine Engineering Faculty, Chabahar Maritime University, Chabahar, IranElec. Eng. Dept., Marine Engineering Faculty, Chabahar Maritime University, Chabahar, IranElectrical Engineering Elec. Eng. Dept., Marine Engineering Faculty, Chabahar Maritime University, Chabahar, IranDepartment, Marine Engineering Faculty, Chabahar Maritime University, Chabahar, IranElec. Eng. Dept., Marine Engineering Faculty, Chabahar Maritime University, Chabahar, Iran<p style="direction: ltr;">Due to the structural complexity of the Persian script and the lack of standardized and reliable datasets, Persian scene text detection and word segmentation in natural scene images captured by conventional cameras remain key challenges in the field of image processing. In this paper, we introduce a comprehensive dataset for Persian text detection, named FATD (Farsi Text Detection Dataset). FATD comprises more than 2,000 diverse images containing texts with various fonts, sizes, orientations, and environmental conditions, covering a wide range of visual complexity. Subsequently, six deep learning models are evaluated and compared under identical conditions on this dataset, including two convolutional neural network (CNN)-based models (YOLOv8 and CRAFT), two transformer-based models (RRDETR and RRBDETR), and two vision-language models (Qwen2.5VL and Florence-2). Experimental results demonstrate that transformer-based models achieve superior accuracy&mdash;up to 65% in H-mean&mdash;at the expense of higher computational cost. In contrast, CNN-based models offer competitive accuracy with notably faster inference speed. Moreover, despite their limited training exposure to Persian text data, the evaluated vision-language models exhibit promising localization performance according to the H-mean metric. Overall, this study provides a valuable benchmark and comparative analysis for advancing Persian scene text detection and highlights the potential of modern vision-language architectures in low-resource languages.</p>http://ijece.org/Article/51191Persian text dataset Scene text detection deep learning models FATD benchmark datasetper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234257266articleMohsen Mahdaviaslmahdaviasl.mohsen@gmail.com1M. Kolahkajmaralkolahkaj@gmail.com2Comp. Eng. Dept., Susangerd Branch, Islamic Azad University, Susangerdو ]قشد<p style="direction: ltr;">Accurate stock price prediction has consistently been one of the fundamental challenges in financial markets, and the development of intelligent models can play a significant role in supporting investors&rsquo; decision-making processes. In this study, a hybrid deep learning&ndash;based framework is proposed for stock price prediction in the Iranian capital market. The proposed approach employs a Convolutional Neural Network (CNN) as the core learning architecture and integrates the Harris Hawks Optimization (HHO) algorithm as a metaheuristic strategy to optimize the model&rsquo;s weights and parameters. The primary objective of this integration is to enhance prediction accuracy while reducing computational complexity through automatic feature extraction within the intermediate layers of the network. The dataset used in this research consists of daily stock information of Bahman Khodro Company from 18/01/1380 to 23/12/1399 (Persian calendar), including variables such as the number of transactions, trading volume, trading value, and prices (previous, opening, closing, final, lowest, and highest). Simulation results demonstrate that the proposed CNN-HHO model outperforms conventional neural network&ndash;based and metaheuristic-based methods, achieving a significantly lower Mean Squared Error (MSE). Overall, the findings indicate that the integration of CNN with the HHO algorithm can serve as an intelligent, accurate, and efficient approach for financial time-series forecasting, providing an effective tool for more informed decision-making in the stock market.</p>http://ijece.org/Article/49951Optimization prediction convolutional neural network deep learningper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234267277articleM. Gholizadehmobin_gholizadeh@comp.iust.ac.ir1N. Mozayenimozayani@iust.ac.ir2Dept. of Comp. Eng., Iran University of Science and Technology, Tehran, IranDept. of Comp. Eng., Iran University of Science and Technology, Tehran, Iran<p style="direction: ltr;">With the rapid growth of computer networks and the increasing complexity of communication infrastructures, the need for efficient resource management and reduced operational and maintenance costs has become a major challenge. Software-Defined Networking (SDN), by decoupling the control and data planes, provides flexibility, scalability, and centralized management. In the control plane, controllers act as the core components responsible for network management and decision-making, and their optimal placement remains a fundamental challenge. The distribution and deployment of controllers directly affect network performance parameters such as delay reduction, load balancing, and stability enhancement. In this paper, a multi-objective machine learning&ndash;based approach is proposed for controller placement in SDN. The problem is formulated with the objectives of minimizing control-plane delay, inter-domain delay, controller load, and the failure cost of network nodes and links, while maximizing network throughput. Considering the computational complexity of the problem, a deep reinforcement learning (DRL) algorithm is employed as the solution approach. Simulation results on multiple network topologies demonstrate that the proposed algorithm improves the control-plane delay, inter-domain delay, and network throughput by approximately 20%, 30%, and 15%, respectively, compared with previous works. Therefore, the proposed approach not only addresses common objectives such as delay minimization and load balancing but also incorporates crucial factors such as stability and fault tolerance into the controller placement decision-making process.</p>http://ijece.org/Article/49934Controller placement software-defined networking (SDN) machine learningper Iranian Research Institute for Electrical Engineeringفصلنامه مهندسی برق و مهندسی کامپيوتر ايران16823745168237452026-03234278284articleMohammad Kahanikahani@mail.um.ac.ir1S. Abrishamis-abrishami@um.ac.ir2A. Nadjaran Toosiadel.toosi@unimelb.edu3Dept. of Comp. Eng., Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, IranDept. of Comp. Eng., Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, IranSchool of Comp. and Info. Systems, The University of Melbourne, Melbourne, Australia<p style="direction: ltr;">Nowadays, the use of serverless computing and Function-as-a-Service (FaaS) applications has been rapidly increasing. This paradigm enables users to deploy their applications without the need to configure infrastructure hosts, while benefiting from advantages such as pay-per-use pricing, flexibility, and automatic scalability of serverless platforms. However, due to the stateless nature of serverless functions, system performance is often limited in scenarios where functions need to interact with each other or when accessing large volumes of data. In common approaches, remote data stores such as Amazon S3 are employed to address this limitation, which introduce significant latency overhead. One solution to mitigate this overhead is the use of caching. Although several studies have explored caching in serverless environments, they generally manage caches either at the host level or at the application level, which does not yield optimal performance. In this paper, we propose a novel framework on an open-source serverless platform that introduces a fine-grained, locality-aware caching system at the function level. This approach not only reduces function response time but also enables more efficient utilization of available resources.</p>http://ijece.org/Article/49722Serverless computing function as a service stateful functions function-level cache origin functions