IEEE Access最新文献

{"title":"A Multi-Criteria Framework for Selecting Machine Learning Techniques for Industrial Fault Prognosis","authors":"Miguel A. De C. Michalski;Carlos A. Murad;Fabio N. Kashiwagi;Gilberto F. M. De Souza;Halley J. B. Da Silva;Hyghor M. Côrtes","doi":"10.1109/ACCESS.2025.3604754","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3604754","url":null,"abstract":"Selecting appropriate Machine Learning (ML) techniques for fault prognosis remains a critical yet often understructured step in developing predictive maintenance strategies for industrial systems. While numerous studies apply ML to estimate Remaining Useful Life (RUL) or forecast failure probabilities, model selection is frequently guided by ad hoc practices or narrow performance metrics, overlooking contextual factors such as data availability, interpretability, automation level, and deployment feasibility. This paper presents a parameterized, multi-criteria decision-making framework to support ML techniques selection in fault prognosis, particularly within energy-related applications. Derived from a structured literature survey, the framework introduces twelve selection parameters, divided into primary requirements and secondary evaluation criteria. These parameters, such as label requirements, model complexity, and transferability, allow users to eliminate unsuitable techniques and rank viable candidates according to application-specific constraints. The framework is applied to a real-world use case involving failure prediction in electrical substations, illustrating how it supports transparent, replicable, and operationally grounded model selection. Results demonstrate the framework’s adaptability to different industrial contexts and its relevance for decision-making in energy systems. By bridging empirical insights with implementation demands, the proposed approach offers a practical tool for aligning ML technique selection with the goals of energy-sector prognostics and maintenance planning.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"154508-154544"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11145754","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Dynamic Malware Behavior Analysis Through Novel Windows Events With Machine Learning","authors":"Göksun Önal;Mesut Güven","doi":"10.1109/ACCESS.2025.3604979","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3604979","url":null,"abstract":"Malware analysis involves studying harmful software to understand its behavior and find ways to detect and prevent it. As cyberattacks become more advanced, this process becomes increasingly important for safeguarding systems and data. Traditional methods in malware analysis often rely on examining the code itself, which can miss malicious actions that only occur during execution. This study addresses this limitation by combining the dynamic observation of malware behavior with an innovative use of Windows Event Logs as input, a detailed system data source. During the study, a secure environment was created to safely execute malware, collect input, and provide valuable information on how malicious software interacts with systems. New methods were developed to extract meaningful information from the logs, then used to train machine-learning models capable of accurately distinguishing malware from legitimate programs. By demonstrating the untapped potential of Windows Event Logs, this study offers new tools to improve real-time malware detection and enhance cybersecurity. On a dataset of approximate 7000 Windows executable file, roughly sixty percent benign and forty percent malware, the log-feature MLP reached 91.2 % accuracy with a 1.6-point standard deviation across five folds, achieved a ROC-AUC of <inline-formula> <tex-math>$0.962~pm ~0.009$ </tex-math></inline-formula> on an unseen hold out set.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"153937-153958"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146719","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Make Large Language Models Efficient: A Review","authors":"Aman Mussa;Zhanseit Tuimebayev;Madina Mansurova","doi":"10.1109/ACCESS.2025.3605110","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3605110","url":null,"abstract":"Large Language Models (LLMs) have achieved remarkable success across a variety of natural language processing tasks, with larger architectures often exhibiting superior performance. This scaling behavior has fueled intense competition in generative AI, supported by projected investments that exceed <inline-formula> <tex-math>${$} $ </tex-math></inline-formula>1 trillion to develop increasingly sophisticated LLMs. This competition has in turn nurtured a vibrant ecosystem, inspiring new open-source models such as DeepSeek, and motivating application developers to harness state-of-the-art LLMs for real-world deployments. However, the extensive memory and computational requirements of large models present serious obstacles for small-medium organizations, leading to significant scalability concerns. This paper offers a comprehensive review of recent techniques to improve LLM efficiency through four categories: parameter-centric, architecture-centric, training-centric and data-centric. For a better understanding of the newcomer’s perspective, it covers the entire lifecycle when developing and deploying LLMs. Thus, this paper is organized around five core tasks: model compression for local deployment, accelerated pre-training to reduce time-to-train, efficient fine-tuning on custom data, optimized inference under resource constraints, and streamlined data preparation. Rather than focusing on broad strategies, we emphasize specialized techniques tailored to each stage of development. By applying targeted optimizations at each phase, the computational overhead can be reduced by 50–95% without compromising the quality of the model, making LLMs more accessible to researchers and practitioners with limited computational resources.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"154466-154490"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146704","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021199","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"KC-UNet: Enhancing U-Net With KAN and CBAM for Medical Image Segmentation","authors":"Jie Xu;Haoqing Gao;Zhifeng Wang","doi":"10.1109/ACCESS.2025.3605148","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3605148","url":null,"abstract":"Medical image segmentation is a critical task in medical image analysis. However, traditional convolutional neural network (CNN) based methods are limited in modeling long-range dependencies, while Transformer-based segmentation models, though effective, suffer from high computational complexity due to quadratic attention operations. To address these challenges, this paper proposes an innovative U-Net variant, KC-UNet, which integrates Kolmogorov-Arnold Networks (KAN) and the Channel-Spatial Attention Module (CBAM). KC-UNet leverages the KAN representation theorem to represent features more efficiently, while CBAM enhances the model’s ability to adaptively capture both spatial and channel-wise dependencies, striking a balance between accuracy and computational efficiency. To validate the effectiveness of CBAM, this paper conducts comprehensive ablation experiments by replacing CBAM with Squeeze-and-Excitation (SE), and Efficient Channel Attention (ECA), as well as removing the attention module entirely. Results demonstrate that CBAM provides the most significant performance improvements in terms of segmentation accuracy, confirming its superior capability in enhancing feature representation. This study evaluates KC-UNet on four widely used benchmark datasets (BUSI, GLAS, CVC, and ISIC2017) and compare it against recent state-of-the-art models such as TransUNet, Swin-unet, and U-KAN. KC-UNet consistently achieves the best performance, with an IoU of 66.60% on BUSI, outperforming Swin-unet by 1.28%, and a Dice score of 80.46%, which improves upon the baseline U-Net by 7.44%. Similar advantages are observed on GLAS and ISIC2017, demonstrating the effectiveness and generalizability of our approach across different modalities. To the best of our knowledge, KC-UNet is the first framework to integrate KAN and CBAM for medical image segmentation.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"153788-153797"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146708","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Systematic Literature Review on Effort Estimation by Software Development Life Cycle Phases","authors":"Sulem Martínez-Aguilar;Ángel J. Sánchez-García;Cuauhtémoc López-Martín;Jorge Octavio Ocharán-Hernández","doi":"10.1109/ACCESS.2025.3605102","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3605102","url":null,"abstract":"Several techniques have been proposed to estimate the total effort of the Software Development Life Cycle (SDLC) rather than by SDLC phase to be performed for an independent team. An accurate effort estimation is needed for software managers to create realistic plans and allocate resources appropriately to independent teams. Therefore, this Systematic Literature Review (SLR), unlike other secondary studies, examines the current state of effort estimation by SDLC phase instead of estimating it across the entire SDLC. In addition, this SLR identifies metaheuristics used for optimizing the parameters of effort estimation models. We searched for studies whose objective has been to propose models for estimating the effort of specific SDLC phases, rather than total SDLC effort. We firstly identified 216 studies, and finally we selected 31 of them published between 2014 and march 2025 in journals and conferences. The majority of the studies investigated effort estimation in the testing and maintenance phases, mainly using Machine Learning (ML) techniques. Functional size and project characteristics were the most common explanatory variables. International Software Benchmarking Standards Group was the predominant dataset for training models and the mean of Absolute Residual was the recommended precision measure. Cross-validation was the most used model validation method. We can conclude that more research is needed on effort estimation by SDLC phases such as requirements specification, design, and construction, as well as to further explore ML and metaheuristics to improve the prediction accuracies of models.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"153340-153358"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146651","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145027971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A New Multi Objective Video Summarization Approach for Video Surveillance Analytics Applications on Smart Cities","authors":"Turan Goktug Altundogan;Mehmet Karaköse;Fatih Mert","doi":"10.1109/ACCESS.2025.3605259","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3605259","url":null,"abstract":"Summarizing surveillance videos used in smart city applications is very important in terms of transaction costs and sustainability. Although there is a considerable amount of literature on video summarization, the methods in the literature for summarizing surveillance videos used in smart cities are few and inadequate. Because, both object and event features of the mentioned video data must be preserved. In this study, we integrated an object-centric and an event-centric summarization method with Apache Kafka for effective summarization of such videos. With the object-centric summarization module of our proposed method, we focused on preserving the statistical and motion features of the objects in the videos. With the event-centric summarization module, we ensured the preservation of abnormal events in the videos. We presented the performance results of both modules and the integrated system in detail with different metrics. Finally, we compared the performances of both modules with the video summarization approaches in the literature based on different metrics. The developed object-centric summarization method preserves the statistical features of the video with a success rate of over 90% and shortens the videos with a rate of over 95%. The developed event-centric summarization approach provides summaries that include abnormal situations found in videos with a success rate of over 95%. The presented comparative results prove that this original method we developed is superior to the studies in the literature in terms of performance and many different evaluations.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"154353-154382"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146661","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021437","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"YOLOv8-ECA-SPD-Lite for Cantilever Bolt Defect Detection in Railway Catenary Systems","authors":"Yapeng Wang;Mengru Chen;Lina Guo;Yi Ma;Min Wu","doi":"10.1109/ACCESS.2025.3604831","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3604831","url":null,"abstract":"To address the challenges of highly variable operating conditions in electrified railway catenary systems, the small target size of cantilever bolt defects, and the large parameter sizes of existing catenary inspection models, this paper proposes an improved lightweight algorithm named You Only Look Once version 8 Enhanced by Efficient Channel Attention, Space-to-Depth Convolution, and Lightweight Optimization via Pruning and Distillation (YOLOv8-ECA-SPD-Lite). The core innovations lie in: 1) Data augmentation on the original images to improve generalization capability; 2) Integration of the Efficient Channel Attention (ECA) module to enhance feature responses in bolt regions for small target detection; 3) The Space-to-Depth convolution (SPD-Conv) module was first applied to cantilever bolt defect detection to minimize feature loss for small targets; and 4) Replacement of the detection head with a decoupled head to enhance defect perception capabilities. Furthermore, applying model pruning and distillation techniques significantly improves model lightweighting, facilitating deployment on embedded onboard systems. Ablation studies validated the effectiveness of each proposed module. Comparative experiments demonstrated that the improved YOLOv8 model outperforms Faster R-CNN, SSD, YOLOv6s, YOLOX-s, YOLOv9s, and the original YOLOv8s across multiple metrics. Specifically, the proposed model achieved a higher mean Average Precision (mAP) than all comparison models. In terms of lightweight design, the improved YOLOv8s model achieves the lowest computational complexity among all compared methods, with only 17.2 GFLOPs. It demonstrates exceptional suitability for mobile deployment. This research holds significant theoretical and practical value, facilitating the transition of railway operation and maintenance to artificial intelligence, enhancing the efficiency of catenary inspection, and meeting the demands of high-density railway operations.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"155167-155180"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146734","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Single-Input Bipolar-Output (SIBO) DC–DC Boost Converter for Solar Generators and On-Chip Power Delivery: Modeling and Experimental Assessment of the So-Called Perilla Converter","authors":"Robert Urbina;Carlos Ivan Paez-Rueda;Germán Yamhure;Manuel R. Pérez;José Vuelvas;Manuel Fernando Párraga Meneses;Abdel Karim Hay Harb;Luis Fernando Melchior Ramirez;Gabriel Perilla Galindo;Arturo Fajardo","doi":"10.1109/ACCESS.2025.3605109","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3605109","url":null,"abstract":"Hybrid power converters that cascade switched-capacitor stages with conventional LC-based DC/DC switching converters have recently emerged as promising solutions for applications requiring a bipolar DC bus, such as photovoltaic systems and in-chip OLED drivers. However, despite clear advantages in efficiency, compactness, and implementation simplicity, their adoption has been limited due to the lack of accurate modeling and experimental validation. Traditional modeling techniques often fail to capture the static and dynamic behavior of these hybrid topologies. Some previous works either neglected capacitive losses or relied on empirical equivalent models without a clearly defined methodology. The Perilla SIBO (Single-Input Bipolar-Output) Boost Converter is a representative example of this new class of hybrid converters, integrating a conventional Single-Inductor Single-Output (SISO) boost stage with a switched-capacitor inverter. In earlier work, a static model for the Perilla Boost converter was derived using the small-ripple approximation and charge balance principles, assuming matched output capacitors and balanced load conditions—but without considering efficiency estimation. This paper extends the previous model to account for unbalanced loading and mismatched output capacitances and introduces a novel analytical expression to estimate converter efficiency. A comprehensive static modeling and experimental evaluation is presented under realistic operating conditions. The model is validated using a prototype operating across an input voltage range of 3 V to 10 V and delivering output power between 1 W and 20 W. The proposed model achieves a mean absolute percentage error of 5.76% and a standard deviation of 5.68%, offering a reliable tool for steady-state performance prediction. The dynamic modeling of the converter, as well as the development of a systematic design methodology, are beyond the scope of this work.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"155411-155424"},"PeriodicalIF":3.6,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146771","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145021321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Analysis of Ansätze Expressibility and Complexity and Their Impact on Classification Accuracy Using QNN and QLSTM","authors":"Sarvapriya M. Tripathi;Himanshu Upadhyay;Jayesh Soni","doi":"10.1109/ACCESS.2025.3604721","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3604721","url":null,"abstract":"Quantum Neural Networks (QNN) and Quantum Long Short-Term Memory (QLSTM) models are emerging as powerful tools in quantum machine learning. The effectiveness of these models is largely governed by the structure of their parameterized quantum circuits, also known as Ansätze. The Expressibility (measure of how well an ansätz can explore Hilbert space), Entanglement (which governs the correlation between states of multiple qubits), and Depth (a measure of how many layers of quantum gates a circuit has) are fundamental to a model’s capacity to learn complex patterns. In this paper, we analyze how varying these attributes for various ansätze influences classification cost and accuracy. We evaluated QNN and QLSTM models across three cybersecurity datasets using ten ansätze with varying expressibility and complexity. Results showed that ansätze with lower expressibility and complexity achieved accuracy and training stability comparable to the more complex ones. Additionally, we found that QNN models consistently offered a better trade-off between accuracy, training time, and computational efficiency compared to QLSTM models.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"152412-152429"},"PeriodicalIF":3.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11145787","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145005444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Channel-Robust Specific Emitter Identification Based on Domain-Adversarial Training of Neural Networks and Multi-Feature Fusion","authors":"Jialong He;Yuelei Xie;Xiangguo Liu","doi":"10.1109/ACCESS.2025.3604428","DOIUrl":"https://doi.org/10.1109/ACCESS.2025.3604428","url":null,"abstract":"To address the significant decline in the accuracy of Specific Emitter Identification(SEI) under wireless channel, we propose a novel method that combines a domain adversarial network with multi-feature fusion(MFF) to extract domain-invariant features of the signal and leverage the complementary nature of signal features extracted from different views. Initially, we employ the IQ Convolutional Neural Network (IQCNN), the Gate Recurrent Unit (GRU), and the stacked Fourier Analysis Networks (SFAN) to directly extract and fuse correlation, temporal, and periodic features from the raw I/Q data. Subsequently, we integrate a Domain-Adversarial Training of Neural Networks (DANN) to eliminate channel features, ultimately enabling SEI under channel interference. The experimental results on the WiFi dataset demonstrate that the MFF network designed in this study achieves an identification accuracy of 97% under Additive White Gaussian Noise(AWGN) channel interference with a signal-to-noise ratio(SNR) of 10dB. Furthermore, the proposed method achieves identification accuracy of 93.8%, 90.3%, and 78.2% under three complex real-world channel interference scenarios, respectively. These findings indicate that the proposed method effectively mitigates channel interference and significantly enhances the robustness of SEI.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"153093-153104"},"PeriodicalIF":3.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11145769","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145027947","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}