https://www.csionjcse.ir/ The CSI Journal on Computer Science and Engineering en daily 1 Wed, 01 Jul 2026 00:00:00 +0330 Wed, 01 Jul 2026 00:00:00 +0330 https://www.csionjcse.ir/article_234693.html Permeability is one of the most important reservoir properties, which indicates the ability of fluids to flow through the pore spaces of the rock. determining permeability in processes such as predicting real reserve, producing and developing oil reservoirs seems essential. In the oil industry, permeability is usually measured using core analysis, well testing, and empirical correlations. The conventional methods of core analysis and well testing are too time-consuming and expensive. also, there data are not provided for every well. On the other hand, empirical correlations are used for special cases and are not accurate for every situation. Due to time-related and financial limitation, developing a method for measuring petrophysical properties such as: permeability based on well logging data (well logging data are available for almost every well) could be significant. An alternative approach for evaluating permeability is the use of artificial intelligence and machine learning tools. In this study, the method of data mining has been applied to calculate reservoir permeability by applying petrophysical data, at first, the data had been normalized and then horizontal and vertical permeability of an Iranian reservoirs were calculated using geophysical data and the methods of multiple layer perceptron Neural Network, PSO and GA. The comparison of these methods showed that combining MLP with either PSO or GA yields the best results. https://www.csionjcse.ir/article_243148.html The rapid advancement of technology underscores universities’ role as innovation and entrepreneurship hubs. With AI increasingly linked to academic environments, there is a growing need to design specialized AI-focused university entrepreneurship ecosystems. This study develops an AI-driven entrepreneurship ecosystem model for Persian Gulf University, Bushehr, using a mixed-methods, semi-quantitative approach. A systematic literature review identified 31 subcomponents, and the views of 17 experts in AI, entrepreneurship, and policy were analyzed via the fuzzy Delphi method. Triangular fuzzy numbers and a five-point scale were applied, with defuzzification using the centroid method and a 0.60 acceptance threshold. Findings reveal five key dimensions: (1) cultural and human, (2) technological and infrastructural, (3) institutional and governance, (4) networked and international, and (5) economic and commercialization, encompassing 31 components. The study offers strategic recommendations to strengthen data governance, local computing infrastructure, and knowledge-based entrepreneurial capacities. https://www.csionjcse.ir/article_240031.html Abstract: The layout design of analog integrated circuits (ICs) is a challenging and time-consuming task, requiring manual effort to extract geometric constraints such as symmetry and proximity. This paper presents a novel approach for automatic layout generation, combining the power of convolutional neural networks (CNNs) with transfer learning, specifically using the pre-trained SqueezeNet model to extract these constraints from schematic images. By applying fine-tuning, the CNN model can effectively identify and extract the necessary geometric relationships, eliminating the need for manual extraction. For the routing process, a multi-objective optimization strategy is employed using the non-dominated sorting genetic algorithm III (NSGA-III), where wire length, via count, and arch segments are minimized. To address the challenge of generating an effective initial population for NSGA-III, we introduce the differential evolution (DE) algorithm as a method for generating high-quality initial solutions, enhancing the convergence speed and solution quality. The proposed methodology is applied to the layout design of a two-stage operational amplifier (op-amp), with simulations conducted using MATLAB and validated in Cadence software on a 0.18μm CMOS process at a 1.8 V supply voltage. The results show that the proposed method signif-icantly outperforms existing techniques in terms of layout efficiency, performance, and automation in the design process. https://www.csionjcse.ir/article_243801.html Natural Language Processing (NLP) is one of the fundamental branches of data science and artificial intelligence, aiming at the automatic analysis, understanding, and generation of human language. Recent advances in this field have played a significant role in improving text analysis systems, information retrieval, machine translation, speech recognition, and intelligent interactive systems. With the rapid growth of textual data and the increasing complexity of linguistic structures, traditional rule-based and simple statistical methods no longer provide sufficient capability for modeling complex and long-range linguistic dependencies. In recent years, deep neural networks, particularly hybrid architectures, have attracted considerable research attention. The combination of Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks enables the simultaneous extraction of local features and long-term temporal dependencies. However, the performance of such models is highly dependent on the proper selection of hyperparameters, and manual tuning is often time-consuming and suboptimal. Therefore, the use of metaheuristic algorithms for automatic hyperparameter optimization has emerged as an effective solution. In this study, a deep learning framework based on a CNN–BiLSTM architecture is presented, in which hyperparameter optimization is performed using the Cuckoo Search algorithm. Experimental results demonstrate that the proposed method achieves an accuracy of 93%, outperforming existing approaches and confirming its effectiveness in analyzing complex textual data while significantly reducing the error rate. https://www.csionjcse.ir/article_242762.html Whole time-series clustering (WTSC) methods often fail to treat sub-patterns as independent entities, limiting their ability to capture diverse behavioral characteristics in time series data. To address this limitation, this study proposes a sliding window-based whole time series clustering method (SW-WTSC) that enhances clustering performance through sub-pattern recognition. The method extracts sub-samples using a sliding window over training data, applies normalization and feature selection, and clusters the resulting sub-samples via k-means. Cluster labels are then aggregated at the sample level, enabling secondary clustering based on the highest similarity among sub-patterns. The proposed approach was evaluated on 85 datasets from the UCR repository. Statistical analysis using the non-parametric Friedman test demonstrated that SW-WTSC significantly outperforms state-of-the-art WTSC methods. These results indicate that incorporating subsequence-based analysis can substantially improve whole time-series clustering accuracy. https://www.csionjcse.ir/article_242763.html Clustering a set of points in the Euclidean plane is a fundamental problem in computational geometry and data analysis, withapplications spanning image segmentation, spatial data mining, and pattern recognition. Minimum Spanning Tree (MST)-based clustering methods offer a geometrically intuitive approach, but suffer from the O(n2 ) cost of constructing the completeEuclidean graph for large point sets. We propose an efficient O(n log n) algorithm for clustering n points in the Euclideanplane. Our method constructs a t-spanner with stretch factor t = 1 + ε for any user-chosen ε > 0 via a Well-Separated PairDecomposition (WSPD), yielding a sparse graph of O(n/ε2 ) edges whose MST approximates the true Euclidean MST within afactor of 1 + ε. For any fixed ε > 0 this is O(n) edges, though the constant grows as ε → 0. Clustering is then performed by anovel local adaptive edge-cutting criterion: an edge (ai , bi ) is removed if its weight exceeds α times λi = max(δ(ai ), δ(bi )),where δ(p) is the local scale at p (approximated in practice by its k-nearest-neighbour distance). This makes the criterion scale-aware and sensitive to local point density. When δ(p) is taken as the FST leaf cell diameter, all quantities needed for cuttingare available as byproducts of the Fair Split Tree construction at no extra asymptotic cost. Experiments on synthetic Gaussiandatasets and standard 2D clustering benchmarks (R15, Aggregation) confirm that the method achieves perfect clustering on well-separated point sets and competitive NMI on denser configurations. https://www.csionjcse.ir/article_246490.html Abstract: Wireless Body Sensor Networks (WBSNs) are increasingly essential in healthcare, enhancing patient comfort and quality of life. These systems support patients, doctors, and medical teams through services such as continuous medical monitoring, medication and health information delivery, memory assistance, home device control, and emergency communication. This paper proposes a novel hybrid method for ECG signal compression, combining the Discrete Wavelet Transform (DWT) with the Invasive Weed Optimization (IWO) algorithm. DWT effectively decomposes the ECG signal into various frequency bands, capturing critical signal features, while IWO optimizes the selection of wavelet coefficients to achieve maximum compression with minimal data loss. The proposed approach was tested on standard ECG datasets, showing improved compression ratios compared to conventional methods. The results confirm that this technique maintains high signal quality, making it highly effective for real-time ECG monitoring in WBSNs. This ensures efficient data transmission and reliable performance in modern healthcare applications. https://www.csionjcse.ir/article_246491.html Recently, research on wireless body sensor networks (WBSNs) has gained significant attention and now plays a crucial role in patient monitoring. Due to power supply limitations, wireless nodes are typically organized into clusters for energy-efficient communication. To this end, clustering-based and grid-based approaches are commonly used. In the first approach, nodes are grouped into clusters in such a way that one sensor node is selected as the cluster head. In contrast, in the grid-based approach, the network is divided into limited virtual grids, usually managed by the base station. The firefly optimization algorithm (FOA), in combination with the K-Means algorithm, is used to optimally select cluster centers. Subsequently, the performance of the proposed segmentation method is evaluated and compared with existing methods. Simulation results show that the network lifetime improves by at least 7% compared to existing approaches. https://www.csionjcse.ir/article_246492.html This paper presents a multilayer encryption framework for enhancing security in cloud computing environments. The proposed approach combines AES-256, Twofish-256, and Serpent-256 symmetric encryption algorithms with RSA-based key exchange and SHA-256 hashing to improve confidentiality, integrity, and secure data transmission. The framework applies sequential encryption layers using independent cryptographic keys in order to increase resistance against unauthorized access and cryptanalytic attacks. Data integrity is verified through comparison of the generated hash values before encryption and after decryption. The proposed model was implemented and evaluated using CloudSim and the TON-IoT dataset under different storage conditions. Experimental results indicate that the proposed multilayer mechanism significantly improves the security margin while introducing only limited computational overhead compared with conventional AES-based approaches. The evaluation also demonstrates acceptable quality-of-service performance for cloud applications and secure IoT environments. https://www.csionjcse.ir/article_247014.html Retrieval-Augmented Generation (RAG) has emerged as an effective approach for improving the accuracy and reliability of Large Language Models (LLMs) by integrating external knowledge retrieval with text generation. However, traditional RAG systems often suffer from limitations such as static retrieval mechanisms, poor context adaptation, and reduced efficiency when handling complex queries. To address these challenges, this paper proposes an Adaptive Hybrid Retrieval-Augmented Generation Framework for context-aware question answering and intelligent response generation.The proposed system combines sparse retrieval techniques using BM25 with dense retrieval through vector-based semantic search to improve document relevance and retrieval accuracy. An adaptive retrieval mechanism dynamically selects suitable retrieval strategies based on query complexity and contextual requirements. Retrieved documents are further processed through a context fusion and ranking module to construct high-quality contextual information for the Large Language Model. The framework supports both offline and online deployment modes, enabling efficient operation in local as well as cloud-based environments.The proposed Hybrid RAG architecture enhances semantic understanding, reduces hallucination, improves response consistency, and provides real-time access to updated knowledge sources. Experimental analysis demonstrates improved retrieval efficiency, response relevance, and scalability compared to traditional retrieval-generation approaches. The system can be effectively applied in intelligent chatbots, domain-specific question answering, educational assistants, healthcare information systems, and enterprise knowledge management applications. https://www.csionjcse.ir/article_247055.html The increasing integration of artificial intelligence into safety-critical and large-scale digital systems challenges conventional cybersecurity risk analysis methods. Traditional approaches struggle to capture AI-specific characteristics such as emergent behavior, vulnerability to adversarial machine learning attacks, and the growing overlap between security, safety, and accountability concerns. At the same time, regulatory initiatives highlight the need for structured, quantitative risk analysis methods for high-risk AI systems. This paper proposes an enhanced cybersecurity risk analysis framework tailored to systems that incorporate AI components. Building on adversarial risk analysis, the framework explicitly models AI-related impacts, learning-based assets, intelligent security and recovery controls, and AI-enabled targeted attacks. System architectures are decomposed into hierarchical blocks, enabling probabilistic simulation of attack entry, propagation, defensive failure, and both local and systemic impacts. Strategic attacker behavior is captured by modeling adversarial decision-making under uncertainty and integrating it into Monte Carlo based risk estimation. The applicability of the framework is demonstrated through a case study involving automated driving systems. The results show that incorporating AI-specific defenses and strategic attacker modeling leads to substantial reductions in tail risk and can induce measurable deterrence effects. The proposed approach supports risk-aware cybersecurity investment decisions and provides an analytical foundation for managing and accessing high-risk AI systems in regulated environments. https://www.csionjcse.ir/article_247126.html Large language models are increasingly used in Persian information systems, yet culturally grounded bias evaluation remains limited. We introduce a natural-scenario counterfactual benchmark constructed through human-directed, ChatGPT-assisted drafting and independently validated by three native Persian speakers. From an initial 100 scenarios, 524 rows, and 424 pairs, validation retained 84 scenarios, 424 rows, and 340 pairs; socioeconomic coverage was reduced to one pair. Five open-weight models were evaluated with deterministic restricted next-token A--E scoring, normal/reverse option orders, scenario-cluster bootstrap intervals, and scenario-level sign-flip tests. Baseline mean absolute gaps were small (0.031--0.043), but Dorna-Llama3-8B and Llama-3.1-8B showed severe option-order sensitivity. Fairness prompting increased gaps for every model and significantly for Qwen2-7B, Qwen2.5-14B, and Qwen3-8B. Qwen2.5-14B and Qwen3-8B achieved the highest BBQ-Fa accuracy (0.817), while Qwen2.5-14B led ISEAR-Fa (0.623). Train-only calibration did not improve held-out gaps. The benchmark, validation documentation, prompts, code, and results are provided in an accompanying anonymous GitHub repository. https://www.csionjcse.ir/article_247013.html This manuscript first explores the application of optimal control theory in the formulation of arbitrage detection in exchange markets, addressing the approaches that price makers should take to derive the fair prices. The primary purpose of this study is to develop a robust optimal control model that enables price makers such as central banks to effectively manage macroeconomic factor exchange rate. We utilize methodologies that integrate stochastic control techniques and numerical simulations. In the second application, we apply the optimal control method to red-black game. To implement the game strategies, a Monte Carlo framework is advised. https://www.csionjcse.ir/article_250354.html Control Chart Pattern Recognition (CCPR) is a fundamental task in Statistical Process Control (SPC) that enables the timely detection of process variations and abnormal operating conditions. Accurate recognition of control chart patterns contributes significantly to quality improvement, fault diagnosis, and process reliability. Although Convolutional Neural Networks (CNNs) have demonstrated outstanding performance in pattern classification problems, designing an effective network architecture remains a challenging task that greatly influences recognition accuracy.This paper proposes a hybrid Genetic Algorithm–Convolutional Neural Network (GA-CNN) model for control chart pattern recognition. In the proposed framework, a Genetic Algorithm (GA) is employed to optimize the convolutional feature extraction stage by determining the most suitable filter configuration for the CNN architecture. The optimized configuration guides the design of a deep multi-layer network consisting of five convolutional layers with hierarchical feature extraction capability. To improve training stability and generalization performance, batch normalization, Rectified Linear Unit (ReLU) activation functions, dropout regularization, and global average pooling (GAP) are incorporated into the architecture.The proposed GA-CNN model is evaluated on a control chart pattern dataset containing various normal and abnormal process conditions. Experimental results demonstrate that the optimized architecture effectively captures discriminative pattern characteristics and achieves a classification accuracy of 99.78%. The obtained results confirm that integrating evolutionary optimization with deep convolutional learning can significantly enhance recognition performance and provide a robust solution for intelligent quality monitoring systems. https://www.csionjcse.ir/article_251919.html Machine learning lifecycle security faces challenges beyond traditional software due to its dependence on data, models, software dependencies, and execution environments. This study proposes a dynamic framework for assessing machine learning-specific threats within enterprise MLSecOps. Security outputs from various tools across the ML lifecycle were collected, normalized, and transformed into unified risk indicators. Risk was assessed using five dimensions: impact severity, likelihood of occurrence, detection difficulty, organizational impact, and attacker accessibility. The resulting scores were dynamically updated according to changes in system components. To determine risk dimension values, a questionnaire was distributed to 17 experts, yielding 12 valid responses. The framework was evaluated using six scenarios: non-TLS communication, data version modification, insecure model files, dependency vulnerabilities, data drift, and prompt injection. Results identified prompt injection (92) and insecure model files (71) as the highest-risk scenarios, demonstrating the framework’s effectiveness in converting heterogeneous security evidence into dynamic risk assessments.