Applied Soft Computing最新文献

A multi-feature collaborative computational neural architecture search method for classifying continuous time series signals 一种用于连续时间序列信号分类的多特征协同计算神经结构搜索方法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-30 DOI: 10.1016/j.asoc.2026.114731

Rui Zhang , Xin-Yu Li , Yan-Jun Zhang

This paper proposes a multi-feature collaborative computational neural architecture search (MFCC-NAS) method for classifying continuous time series signals. It is designed to enhance the value density of the signals and mitigate the subjective nature of the design of the model architecture while reducing the computational cost of assessing its performance. We first design a representation of sample richness and a metric of their contributions to enhance the richness of the features in them, so that they provide a basis of dense and high-value data. Following this, we develop a MFCC-NAS method that can efficiently construct a model to classify continuous time series signals by using a search space designed for global–local feature separation based on cells, an architecture search strategy for the collaborative computation of global–local features, and a low-cost and robust cascading strategy to evaluate the performance of the architecture. Finally, we design a multi-domain collaborative fusion mechanism to fully integrate convolutional visual features from different spatial domains and obtain a comprehensive representation of the features of the samples. We tested the proposed method through comparative generalization experiments on a dataset of welding defects and the Yaseen dataset. After respective search times of 1.52 h and 1.21 h on these datasets, our model achieved an accuracy of classification of over 98 % on both. Furthermore, the resulting model maintains a compact parameter size and short inference time. These results collectively demonstrate the effectiveness and strong generalization capability of the proposed MFCC-NAS method.

提出了一种用于连续时间序列信号分类的多特征协同计算神经结构搜索（mfc - nas）方法。它旨在增强信号的值密度，减轻模型体系结构设计的主观性，同时降低评估其性能的计算成本。我们首先设计了样本丰富度的表示及其贡献度量，以增强样本中特征的丰富度，从而为密集和高价值的数据提供基础。在此基础上，我们开发了一种mfc - nas方法，该方法通过基于单元的全局-局部特征分离搜索空间，全局-局部特征协同计算的架构搜索策略以及低成本和鲁棒的级联策略来评估架构的性能，可以有效地构建连续时间序列信号分类模型。最后，设计了一种多域协同融合机制，将不同空间域的卷积视觉特征充分融合，得到样本特征的综合表示。通过焊接缺陷数据集和Yaseen数据集的对比泛化实验，验证了该方法的有效性。在这些数据集上分别搜索1.52 h和1.21 h后，我们的模型在两个数据集上都实现了超过98 %的分类准确率。此外，所得到的模型保持了紧凑的参数大小和较短的推理时间。这些结果共同证明了所提出的mfc - nas方法的有效性和较强的泛化能力。

{"title":"A multi-feature collaborative computational neural architecture search method for classifying continuous time series signals","authors":"Rui Zhang , Xin-Yu Li , Yan-Jun Zhang","doi":"10.1016/j.asoc.2026.114731","DOIUrl":"10.1016/j.asoc.2026.114731","url":null,"abstract":"<div><div>This paper proposes a multi-feature collaborative computational neural architecture search (MFCC-NAS) method for classifying continuous time series signals. It is designed to enhance the value density of the signals and mitigate the subjective nature of the design of the model architecture while reducing the computational cost of assessing its performance. We first design a representation of sample richness and a metric of their contributions to enhance the richness of the features in them, so that they provide a basis of dense and high-value data. Following this, we develop a MFCC-NAS method that can efficiently construct a model to classify continuous time series signals by using a search space designed for global–local feature separation based on cells, an architecture search strategy for the collaborative computation of global–local features, and a low-cost and robust cascading strategy to evaluate the performance of the architecture. Finally, we design a multi-domain collaborative fusion mechanism to fully integrate convolutional visual features from different spatial domains and obtain a comprehensive representation of the features of the samples. We tested the proposed method through comparative generalization experiments on a dataset of welding defects and the Yaseen dataset. After respective search times of 1.52 h and 1.21 h on these datasets, our model achieved an accuracy of classification of over 98 % on both. Furthermore, the resulting model maintains a compact parameter size and short inference time. These results collectively demonstrate the effectiveness and strong generalization capability of the proposed MFCC-NAS method.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114731"},"PeriodicalIF":6.6,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146096115","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Diagnosis of autism disorder from rs-fMRI brain images through hierarchical YOLO and mechanism of attention (HYMA) 基于层次YOLO和注意机制的rs-fMRI脑图像诊断自闭症障碍

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-29 DOI: 10.1016/j.asoc.2026.114655

Saba Gholami , Sara Motamed , Elham Askari

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by atypical communication among brain regions. Early and accurate diagnosis of ASD can significantly improve the effectiveness of behavioral interventions. In this study, we propose a novel deep learning framework called HYMA (Hierarchical YOLO with Mechanism of Attention) for classifying ASD and control subjects using resting-state functional MRI (rs-fMRI) data. The proposed model integrates the YOLO detection architecture with a Bottleneck Attention Module (BAM) to enhance spatial–temporal feature extraction, and employs both dynamic (expert combination) and static (Naïve Bayes) ensemble classifiers for final decision fusion. To address the limited-sample challenge in medical imaging, we adopted extensive data augmentation and subject-wise validation. Experimental results on the ABIDE I and II datasets demonstrate that HYMA achieves superior performance, reaching an accuracy of 99.3%, precision of 98.9%, recall of 99.1%, and F1-score of 99.0%, outperforming existing state-of-the-art methods. These results indicate that the proposed attention-enhanced YOLO ensemble framework provides a robust and generalizable approach for rs-fMRI-based ASD diagnosis.

自闭症谱系障碍（ASD）是一种复杂的神经发育疾病，其特征是大脑区域之间的非典型交流。ASD的早期准确诊断可以显著提高行为干预的有效性。在这项研究中，我们提出了一个新的深度学习框架HYMA (Hierarchical YOLO with Mechanism of Attention)，用于使用静息状态功能MRI （rs-fMRI）数据对ASD和对照受试者进行分类。该模型将YOLO检测体系结构与瓶颈注意模块（BAM）相结合，增强了时空特征提取，并采用动态（专家组合）和静态（Naïve贝叶斯）集成分类器进行最终决策融合。为了解决医学成像中样本有限的挑战，我们采用了广泛的数据增强和主体验证。实验结果表明，HYMA的准确率为99.3%，精密度为98.9%，召回率为99.1%，f1分数为99.0%，优于现有的最先进的方法。这些结果表明，所提出的注意力增强YOLO集成框架为基于rs- fmri的ASD诊断提供了一种强大且可推广的方法。

{"title":"Diagnosis of autism disorder from rs-fMRI brain images through hierarchical YOLO and mechanism of attention (HYMA)","authors":"Saba Gholami , Sara Motamed , Elham Askari","doi":"10.1016/j.asoc.2026.114655","DOIUrl":"10.1016/j.asoc.2026.114655","url":null,"abstract":"<div><div>Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by atypical communication among brain regions. Early and accurate diagnosis of ASD can significantly improve the effectiveness of behavioral interventions. In this study, we propose a novel deep learning framework called HYMA (Hierarchical YOLO with Mechanism of Attention) for classifying ASD and control subjects using resting-state functional MRI (rs-fMRI) data. The proposed model integrates the YOLO detection architecture with a Bottleneck Attention Module (BAM) to enhance spatial–temporal feature extraction, and employs both dynamic (expert combination) and static (Naïve Bayes) ensemble classifiers for final decision fusion. To address the limited-sample challenge in medical imaging, we adopted extensive data augmentation and subject-wise validation. Experimental results on the ABIDE I and II datasets demonstrate that HYMA achieves superior performance, reaching an accuracy of 99.3%, precision of 98.9%, recall of 99.1%, and F1-score of 99.0%, outperforming existing state-of-the-art methods. These results indicate that the proposed attention-enhanced YOLO ensemble framework provides a robust and generalizable approach for rs-fMRI-based ASD diagnosis.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114655"},"PeriodicalIF":6.6,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Triangle centroid search algorithm: A reliable optimization technique for modeling photovoltaic systems 三角质心搜索算法：一种可靠的光伏系统建模优化技术

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-29 DOI: 10.1016/j.asoc.2026.114694

Haonan Zhao, Jing Li

The rapid global expansion of photovoltaic (PV) installations has heightened the importance of PV system design, control and monitoring. Precise parameter identification is critical for PV systems, yet existing metaheuristics often suffer from high variance and unstable convergence due to model complexity. To address this, we propose the triangle centroid search algorithm (TCSA), a novel geometry-inspired metaheuristic. TCSA dynamically constructs triangular subsystems and employs a fitness-weighted centroid to balance exploration and exploitation. By coordinating both internal and external learning strategies, the TCSA enhances convergence speed and enables the population to break free from local optima. Evaluations on single-, double- and triple-diode models demonstrate the superiority of TCSA. Notably, on the complex DDM and TDM, it reduces the root mean square error (RMSE) to 9.8248E-04 while achieving a remarkable standard deviation in the order of 10E-17. Furthermore, extensive assessments on the IEEE CEC2020 benchmark functions and real-world constrained optimization suites validate the algorithm’s scalability and its potential for general engineering tasks. These results confirm that TCSA provides a dynamic PV model optimization framework with high precision and stability.

随着全球光伏发电装机的迅速扩张，光伏系统设计、控制和监测的重要性日益凸显。精确的参数辨识对光伏系统至关重要，但由于模型的复杂性，现有的元启发式算法往往存在高方差和不稳定收敛的问题。为了解决这个问题，我们提出了三角形质心搜索算法（TCSA），这是一种新的几何启发的元启发式算法。TCSA动态构建三角形子系统，并采用适应度加权质心平衡勘探与开发。通过协调内部和外部学习策略，TCSA提高了收敛速度，使群体能够摆脱局部最优。对单二极管、双二极管和三二极管模型的评估表明了TCSA的优越性。值得注意的是，在复杂的DDM和TDM上，它将均方根误差（RMSE）降低到9.8248E-04，同时实现了10E-17量级的显著标准偏差。此外，对IEEE CEC2020基准函数和现实世界约束优化套件的广泛评估验证了该算法的可扩展性及其在一般工程任务中的潜力。这些结果证实了TCSA提供了一个高精度、高稳定性的动态PV模型优化框架。

{"title":"Triangle centroid search algorithm: A reliable optimization technique for modeling photovoltaic systems","authors":"Haonan Zhao, Jing Li","doi":"10.1016/j.asoc.2026.114694","DOIUrl":"10.1016/j.asoc.2026.114694","url":null,"abstract":"<div><div>The rapid global expansion of photovoltaic (PV) installations has heightened the importance of PV system design, control and monitoring. Precise parameter identification is critical for PV systems, yet existing metaheuristics often suffer from high variance and unstable convergence due to model complexity. To address this, we propose the triangle centroid search algorithm (TCSA), a novel geometry-inspired metaheuristic. TCSA dynamically constructs triangular subsystems and employs a fitness-weighted centroid to balance exploration and exploitation. By coordinating both internal and external learning strategies, the TCSA enhances convergence speed and enables the population to break free from local optima. Evaluations on single-, double- and triple-diode models demonstrate the superiority of TCSA. Notably, on the complex DDM and TDM, it reduces the root mean square error (RMSE) to 9.8248E-04 while achieving a remarkable standard deviation in the order of 10E-17. Furthermore, extensive assessments on the IEEE CEC2020 benchmark functions and real-world constrained optimization suites validate the algorithm’s scalability and its potential for general engineering tasks. These results confirm that TCSA provides a dynamic PV model optimization framework with high precision and stability.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114694"},"PeriodicalIF":6.6,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146096116","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Green AI driven quantile boosting ensembles for sustainable real estate valuation 绿色人工智能驱动的可持续房地产估值分位数提升组合

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-29 DOI: 10.1016/j.asoc.2026.114616

Muhammet Ali Kadıoğlu , Mehmet Yavuz Yağcı

In today’s volatile economic environment, the accurate estimation of apartment prices is crucial for maintaining financial predictability. Nevertheless, the development of high-performing artificial intelligence (AI) models often entails substantial computational costs, which, in turn, raise critical concerns regarding environmental sustainability, economic viability, and equitable accessibility. This study addresses these challenges by proposing a comprehensive, energy-efficient framework for real estate price estimation grounded in the principles of Green AI. Utilizing a dataset of over 13 million geolocated listings in Turkey (January 2018–September 2024), we implement an end-to-end machine learning pipeline, including data preparation, outlier detection, quantile-based ensemble modeling, hyperparameter tuning using Optuna, and model evaluation. The performance of models—LightGBM, Gradient Boosting Regressor, XGBoost, and Random Forest—is assessed through a dual lens of estimation accuracy and environmental efficiency, incorporating metrics such as training time, energy consumption, and prediction latency. Results show that LightGBM offers the best balance between accuracy and resource efficiency, making it the recommended baseline for large-scale or real-time valuation systems, while GBR and XGBoost may be used in R&D settings under energy constraints. The paper also highlights the importance of integrating dynamic market indicators and climate risk variables to enhance model robustness and sustainability. Overall, this research advances a holistic framework for responsible and scalable real estate valuation, contributing to the growing field of sustainable AI deployment in high-impact sectors.

在当今动荡的经济环境中，准确估计公寓价格对于保持财务可预测性至关重要。然而，高性能人工智能（AI）模型的开发往往需要大量的计算成本，这反过来又引发了对环境可持续性、经济可行性和公平可及性的关键担忧。本研究通过提出一个基于绿色人工智能原则的全面、节能的房地产价格估算框架来解决这些挑战。利用土耳其超过1300万个地理位置列表的数据集（2018年1月至2024年9月），我们实现了端到端的机器学习管道，包括数据准备、异常值检测、基于分位数的集成建模、使用Optuna进行超参数调优和模型评估。lightgbm、Gradient Boosting Regressor、XGBoost和Random forest等模型的性能通过估计精度和环境效率的双重视角进行评估，并结合训练时间、能耗和预测延迟等指标。结果表明，LightGBM在准确性和资源效率之间提供了最佳平衡，使其成为大规模或实时评估系统的推荐基准，而GBR和XGBoost可用于能源限制的研发环境。本文还强调了整合动态市场指标和气候风险变量以增强模型稳健性和可持续性的重要性。总体而言，本研究提出了一个负责任和可扩展的房地产估值的整体框架，有助于在高影响力行业中可持续部署人工智能。

{"title":"Green AI driven quantile boosting ensembles for sustainable real estate valuation","authors":"Muhammet Ali Kadıoğlu , Mehmet Yavuz Yağcı","doi":"10.1016/j.asoc.2026.114616","DOIUrl":"10.1016/j.asoc.2026.114616","url":null,"abstract":"<div><div>In today’s volatile economic environment, the accurate estimation of apartment prices is crucial for maintaining financial predictability. Nevertheless, the development of high-performing artificial intelligence (AI) models often entails substantial computational costs, which, in turn, raise critical concerns regarding environmental sustainability, economic viability, and equitable accessibility. This study addresses these challenges by proposing a comprehensive, energy-efficient framework for real estate price estimation grounded in the principles of Green AI. Utilizing a dataset of over 13 million geolocated listings in Turkey (January 2018–September 2024), we implement an end-to-end machine learning pipeline, including data preparation, outlier detection, quantile-based ensemble modeling, hyperparameter tuning using Optuna, and model evaluation. The performance of models—LightGBM, Gradient Boosting Regressor, XGBoost, and Random Forest—is assessed through a dual lens of estimation accuracy and environmental efficiency, incorporating metrics such as training time, energy consumption, and prediction latency. Results show that LightGBM offers the best balance between accuracy and resource efficiency, making it the recommended baseline for large-scale or real-time valuation systems, while GBR and XGBoost may be used in R&D settings under energy constraints. The paper also highlights the importance of integrating dynamic market indicators and climate risk variables to enhance model robustness and sustainability. Overall, this research advances a holistic framework for responsible and scalable real estate valuation, contributing to the growing field of sustainable AI deployment in high-impact sectors.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114616"},"PeriodicalIF":6.6,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080297","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A multi-modal brain image fusion technique using nakagami imaging and intuitionistic fuzzy sets 基于nakagami成像和直觉模糊集的多模态脑图像融合技术

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.asoc.2026.114706

K.G. Lavanya, P. Dhanalakshmi, M. Nandhini

Medical image fusion is essential for consolidating images from diverse modalities into a single image, offering comprehensive information for diagnosis and analysis. With the rapid evolution of brain imaging technologies, researchers increasingly focus on refining fuzzy fusion techniques to utilize the full potential of these modalities. However, existing fuzzy fusion approaches often face three key challenges: difficulty in enhancing low-contrast images, uncertainty in selecting appropriate fuzzy membership functions, and redundancy in preserving important features. To address these shortcomings, a Nakagami distribution and intuitionistic fuzzy set-based multi-modal brain image fusion (NIFMBF) framework is proposed. This method integrates three core innovations: First, Nakagami imaging (NI) is employed to enhance low-contrast areas and reveal imperceptible lesions; second, a novel intuitionistic fuzzy generator (IFG) is designed to transform NI outputs into intuitionistic fuzzy images that effectively handle vagueness and enhance fine structural details; third, gray-level co-occurrence matrix (GLCM)-based contrast feature extraction guides the fusion process, enabling the retention of crucial information while reducing redundant data. Extensive experiments conducted on three benchmark datasets and compared against five state-of-the-art fusion techniques demonstrate the superior performance of the proposed NIFMBF method. Quantitatively, the method achieves an average improvement of 12–20% for every considered fusion metric. Additionally, the proposed framework reduces computational time by nearly 30% highlighting its efficiency. These findings, validated by both qualitative and quantitative evaluations, underscore the efficacy and potential of NIFMBF for medical image fusion.

医学图像融合对于将不同模式的图像合并为单一图像，为诊断和分析提供全面的信息至关重要。随着脑成像技术的快速发展，研究人员越来越关注模糊融合技术的改进，以充分利用这些模式的潜力。然而，现有的模糊融合方法往往面临三个关键挑战：难以增强低对比度图像，选择合适的模糊隶属函数的不确定性，以及保留重要特征的冗余。针对这些不足，提出了一种基于Nakagami分布和直觉模糊集的多模态脑图像融合（NIFMBF）框架。该方法整合了三个核心创新：首先，利用中川成像（Nakagami imaging， NI）增强低对比度区域，显示难以察觉的病变；其次，设计了一种新的直觉模糊生成器（IFG），将NI输出转化为直觉模糊图像，有效地处理模糊性并增强精细结构细节；第三，基于灰度共生矩阵（GLCM）的对比度特征提取指导融合过程，在减少冗余数据的同时保留关键信息。在三个基准数据集上进行了大量实验，并与五种最先进的融合技术进行了比较，证明了所提出的NIFMBF方法的优越性能。在定量上，该方法对每个考虑的融合度量实现了12-20%的平均改进。此外，所提出的框架减少了近30%的计算时间，突出了其效率。这些发现得到了定性和定量评价的验证，强调了NIFMBF在医学图像融合方面的功效和潜力。

{"title":"A multi-modal brain image fusion technique using nakagami imaging and intuitionistic fuzzy sets","authors":"K.G. Lavanya, P. Dhanalakshmi, M. Nandhini","doi":"10.1016/j.asoc.2026.114706","DOIUrl":"10.1016/j.asoc.2026.114706","url":null,"abstract":"<div><div>Medical image fusion is essential for consolidating images from diverse modalities into a single image, offering comprehensive information for diagnosis and analysis. With the rapid evolution of brain imaging technologies, researchers increasingly focus on refining fuzzy fusion techniques to utilize the full potential of these modalities. However, existing fuzzy fusion approaches often face three key challenges: difficulty in enhancing low-contrast images, uncertainty in selecting appropriate fuzzy membership functions, and redundancy in preserving important features. To address these shortcomings, a Nakagami distribution and intuitionistic fuzzy set-based multi-modal brain image fusion (NIFMBF) framework is proposed. This method integrates three core innovations: First, Nakagami imaging (NI) is employed to enhance low-contrast areas and reveal imperceptible lesions; second, a novel intuitionistic fuzzy generator (IFG) is designed to transform NI outputs into intuitionistic fuzzy images that effectively handle vagueness and enhance fine structural details; third, gray-level co-occurrence matrix (GLCM)-based contrast feature extraction guides the fusion process, enabling the retention of crucial information while reducing redundant data. Extensive experiments conducted on three benchmark datasets and compared against five state-of-the-art fusion techniques demonstrate the superior performance of the proposed NIFMBF method. Quantitatively, the method achieves an average improvement of 12–20% for every considered fusion metric. Additionally, the proposed framework reduces computational time by nearly 30% highlighting its efficiency. These findings, validated by both qualitative and quantitative evaluations, underscore the efficacy and potential of NIFMBF for medical image fusion.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114706"},"PeriodicalIF":6.6,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080294","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Relation-aware heterogeneous graph network multi-modal predictive modeling of stock movements 关系感知的异质图网络股票走势多模态预测建模

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.asoc.2026.114703

Abdullah Ali Salamai

The predictive modeling of movements in stock price information has been common but a challenging task for achieving sustainable management of stock marketplaces. This can be attributed to the importance of this information in evading market risks and improving financing decisions. This task becomes more challenging because of the availability of different modalities of stock information (i.e., stock prices, tweets, events, charts, etc.) and the presence of lead-lag relationships. In this regard, graph neural networks (GNNs) have recently achieved great improvements in providing effective modeling and analysis of stock relationships. However, multi-modal stock data evolve with dynamism in the form of heterogeneous graph topologies that limit the representation power of the existing GNNs. To this end, this study presents a novel graph intelligence framework called relation-aware heterogeneous graph network (RHGN) for efficient prediction of stock movements from multi-modal information by learning different forms of relational knowledge from heterogeneous graphs constructed for every trading day. In particular, the relational graph convolutions are presented to extract distinctive node-wise relational knowledge from the subgraph of each type of relationship. Then, the multi-relation knowledge passing module is designed to empower the network to model the dynamicity of relational knowledge throughout various types of relationships. Simultaneously, the edge relations are encoded into trainable temporal representations, from which edgewise relational knowledge is learned by capturing the contextual interactions among stock entities. In RHGN, an adversarial graph learning mechanism is introduced to adaptively augment and perturb the node representations based on the gradient information to make the model robust against small oscillations in input data. Experimentations on four public multi-modal stock movement datasets have validated the efficiency of the RHGN over the state-of-the-art.

股票价格信息运动的预测建模已经很普遍，但对于实现股票市场的可持续管理是一项具有挑战性的任务。这可以归因于这些信息在规避市场风险和改善融资决策方面的重要性。由于股票信息的不同形式（即股票价格、推文、事件、图表等）的可用性以及领先-滞后关系的存在，这项任务变得更具挑战性。在这方面，图神经网络（gnn）最近在提供有效的股票关系建模和分析方面取得了很大的进步。然而，多模态库存数据以异构图拓扑的形式动态演化，限制了现有gnn的表示能力。为此，本研究提出了一种新的图智能框架，称为关系感知异构图网络（RHGN），通过从每个交易日构建的异构图中学习不同形式的关系知识，从多模态信息中有效预测股票走势。特别地，提出了关系图卷积，从每种关系类型的子图中提取独特的节点智能关系知识。然后，设计了多关系知识传递模块，使网络能够在各种类型的关系中对关系知识进行动态建模。同时，将边缘关系编码为可训练的时态表示，通过捕获库存实体之间的上下文交互来学习边缘关系知识。在RHGN中，引入了一种对抗图学习机制来自适应地增强和扰动基于梯度信息的节点表示，使模型对输入数据的小振荡具有鲁棒性。在四个公共多模态家畜运动数据集上的实验验证了RHGN的效率。

{"title":"Relation-aware heterogeneous graph network multi-modal predictive modeling of stock movements","authors":"Abdullah Ali Salamai","doi":"10.1016/j.asoc.2026.114703","DOIUrl":"10.1016/j.asoc.2026.114703","url":null,"abstract":"<div><div>The predictive modeling of movements in stock price information has been common but a challenging task for achieving sustainable management of stock marketplaces. This can be attributed to the importance of this information in evading market risks and improving financing decisions. This task becomes more challenging because of the availability of different modalities of stock information (i.e., stock prices, tweets, events, charts, etc.) and the presence of lead-lag relationships. In this regard, graph neural networks (GNNs) have recently achieved great improvements in providing effective modeling and analysis of stock relationships. However, multi-modal stock data evolve with dynamism in the form of heterogeneous graph topologies that limit the representation power of the existing GNNs. To this end, this study presents a novel graph intelligence framework called relation-aware heterogeneous graph network (RHGN) for efficient prediction of stock movements from multi-modal information by learning different forms of relational knowledge from heterogeneous graphs constructed for every trading day. In particular, the relational graph convolutions are presented to extract distinctive node-wise relational knowledge from the subgraph of each type of relationship. Then, the multi-relation knowledge passing module is designed to empower the network to model the dynamicity of relational knowledge throughout various types of relationships. Simultaneously, the edge relations are encoded into trainable temporal representations, from which edgewise relational knowledge is learned by capturing the contextual interactions among stock entities. In RHGN, an adversarial graph learning mechanism is introduced to adaptively augment and perturb the node representations based on the gradient information to make the model robust against small oscillations in input data. Experimentations on four public multi-modal stock movement datasets have validated the efficiency of the RHGN over the state-of-the-art.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114703"},"PeriodicalIF":6.6,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A novel graph neural network-based approach for android malware detection 一种新的基于图神经网络的android恶意软件检测方法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.asoc.2026.114689

Donghai Tian , Zhanyun Niu , Tao Leng , Jiaqing Jiang , Pengxuan Chen , Changzhen Hu , Chong Yuan , Ruilong Deng

With the rapid development of the Internet of Things (IoT) and the Internet of Vehicles (IoV), smartphones have evolved into central hubs for connecting users with various smart devices, making their security increasingly vital. However, the growing prevalence of malicious mobile applications poses significant threats to user privacy and digital assets. Existing machine learning-based mobile malware detection methods often face limitations in terms of robustness and interpretability. To address these challenges, we propose GRED (GNN-based Robust and Explainable Malware Detection), an Android malware detection model that leverages graph neural networks to precisely identify malicious behavior while enhancing both robustness and interpretability. GRED refines Android function call graphs, extracts semantic and structural API features, and utilizes a Top-K-based GNN architecture for effective malware detection. Additionally, it offers multi-perspective interpretability analysis to support an in-depth understanding of detection results. Extensive experiments conducted on two large datasets demonstrate that GRED achieves superior performance compared to existing methods. The interpretability module effectively pinpoints malicious behaviors, thereby assisting security analysts in subsequent investigation and mitigation efforts.

随着物联网（IoT）和车联网（IoV）的快速发展，智能手机已经发展成为连接用户与各种智能设备的中心枢纽，其安全性变得越来越重要。然而，恶意移动应用程序的日益流行对用户隐私和数字资产构成了重大威胁。现有的基于机器学习的移动恶意软件检测方法在鲁棒性和可解释性方面往往面临局限性。为了应对这些挑战，我们提出了GRED（基于gnn的鲁棒和可解释恶意软件检测），这是一种利用图神经网络精确识别恶意行为的Android恶意软件检测模型，同时增强了鲁棒性和可解释性。GRED细化Android函数调用图，提取语义和结构API特征，并利用基于top的GNN架构进行有效的恶意软件检测。此外，它还提供了多角度的可解释性分析，以支持对检测结果的深入理解。在两个大型数据集上进行的大量实验表明，与现有方法相比，GRED取得了更好的性能。可解释性模块有效地查明恶意行为，从而协助安全分析人员进行后续调查和缓解工作。

{"title":"A novel graph neural network-based approach for android malware detection","authors":"Donghai Tian , Zhanyun Niu , Tao Leng , Jiaqing Jiang , Pengxuan Chen , Changzhen Hu , Chong Yuan , Ruilong Deng","doi":"10.1016/j.asoc.2026.114689","DOIUrl":"10.1016/j.asoc.2026.114689","url":null,"abstract":"<div><div>With the rapid development of the Internet of Things (IoT) and the Internet of Vehicles (IoV), smartphones have evolved into central hubs for connecting users with various smart devices, making their security increasingly vital. However, the growing prevalence of malicious mobile applications poses significant threats to user privacy and digital assets. Existing machine learning-based mobile malware detection methods often face limitations in terms of robustness and interpretability. To address these challenges, we propose GRED (GNN-based Robust and Explainable Malware Detection), an Android malware detection model that leverages graph neural networks to precisely identify malicious behavior while enhancing both robustness and interpretability. GRED refines Android function call graphs, extracts semantic and structural API features, and utilizes a Top-K-based GNN architecture for effective malware detection. Additionally, it offers multi-perspective interpretability analysis to support an in-depth understanding of detection results. Extensive experiments conducted on two large datasets demonstrate that GRED achieves superior performance compared to existing methods. The interpretability module effectively pinpoints malicious behaviors, thereby assisting security analysts in subsequent investigation and mitigation efforts.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114689"},"PeriodicalIF":6.6,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146096145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multiscale personalized federated load forecasting via enhanced ensemble empirical mode decomposition and Bayesian long short-term memory network 基于增强集成经验模态分解和贝叶斯长短期记忆网络的多尺度个性化联邦负荷预测

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.asoc.2026.114707

Yang Liu , Yingchun Wang

To enhance the personalized modeling capability of federated learning under non-independent and identically distributed (Non-IID) conditions, this study proposes a multiscale personalized federated framework for load forecasting. The framework employs a dual-model architecture that achieves “global trend sharing” and “local feature modeling,” effectively balancing collaboration and personalization across heterogeneous clients. At the data level, an enhanced ensemble empirical mode decomposition (E-EEMD) with paired noise cancellation and adaptive amplitude control is developed to suppress mode mixing and residual noise. A three-dimensional classification criterion integrating zero-crossing rate, Pearson correlation, and variance contribution enables adaptive component identification and multiscale feature extraction. At the model level, an enhanced Bayesian long short-term memory network (B-LSTM) incorporates temporal representation and uncertainty quantification, supporting confidence-weighted aggregation within federated learning. Experimental results across multiple regions verify that the proposed framework consistently outperforms conventional approaches in forecasting accuracy, robustness, and privacy preservation.

为了增强联邦学习在非独立同分布（非iid）条件下的个性化建模能力，本研究提出了一种多尺度的负荷预测个性化联邦框架。该框架采用双模型架构，实现了“全局趋势共享”和“局部特征建模”，有效地平衡了跨异构客户端的协作和个性化。在数据层面，提出了一种增强的集成经验模态分解（E-EEMD）方法，结合配对噪声抵消和自适应幅度控制来抑制模态混合和残余噪声。基于零交叉率、Pearson相关性和方差贡献的三维分类标准实现了自适应成分识别和多尺度特征提取。在模型层面，增强的贝叶斯长短期记忆网络（B-LSTM）结合了时间表征和不确定性量化，支持联邦学习中的置信度加权聚合。跨多个区域的实验结果验证了所提出的框架在预测准确性、鲁棒性和隐私保护方面始终优于传统方法。

{"title":"Multiscale personalized federated load forecasting via enhanced ensemble empirical mode decomposition and Bayesian long short-term memory network","authors":"Yang Liu , Yingchun Wang","doi":"10.1016/j.asoc.2026.114707","DOIUrl":"10.1016/j.asoc.2026.114707","url":null,"abstract":"<div><div>To enhance the personalized modeling capability of federated learning under non-independent and identically distributed (Non-IID) conditions, this study proposes a multiscale personalized federated framework for load forecasting. The framework employs a dual-model architecture that achieves “global trend sharing” and “local feature modeling,” effectively balancing collaboration and personalization across heterogeneous clients. At the data level, an enhanced ensemble empirical mode decomposition (E-EEMD) with paired noise cancellation and adaptive amplitude control is developed to suppress mode mixing and residual noise. A three-dimensional classification criterion integrating zero-crossing rate, Pearson correlation, and variance contribution enables adaptive component identification and multiscale feature extraction. At the model level, an enhanced Bayesian long short-term memory network (B-LSTM) incorporates temporal representation and uncertainty quantification, supporting confidence-weighted aggregation within federated learning. Experimental results across multiple regions verify that the proposed framework consistently outperforms conventional approaches in forecasting accuracy, robustness, and privacy preservation.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114707"},"PeriodicalIF":6.6,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080199","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Optimization method of process parameters based on the constraints of surface quality and geometric feature of elements during copying turning 仿形车削过程中基于表面质量和零件几何特征约束的工艺参数优化方法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-27 DOI: 10.1016/j.asoc.2026.114711

Zhiheng Chen , Ning Li , Shiying Tu , Kaimao Zeng , Juan Lu , Dan Chen

Machining quality and performance are key indicators of CNC machine tool. To enhance the machining of non-circular curves and rotating parts on mid- to low-end CNC machines, this study proposes an optimization method for contour turning parameters based on surface quality and geometric feature constraints. The method integrates trajectory and parameter optimization. First, the Double-Heads Snake (DHS) algorithm generates a smooth G-code tool path comprising straight lines, convex arcs, and concave arcs with varying curvatures. Multi-objective optimization models are then developed for these elements to enhance machining quality and performance stability. Using Back Propagation Neural Networks (BPNN), the models predict surface roughness (Ra) and cutting force (F) based on process parameters and curvature radius, analyzing their influence mechanisms. An improved African Vulture multi-objective optimization algorithm (MOAVOA_improve) is proposed and validated against other algorithms using CEC2009 benchmarks, demonstrating superior performance. Surface quality and geometric constraints guide tool path segmentation, merging adjacent path segments with minimal curvature impact on Ra into single regions. Optimized process parameters are then determined for each region based on curvature and multi-objective solutions. The method is applied to non-circular rotating workpieces and compared with Mastercam and experience-based parameters. Results show significant improvements in machining quality, validating the method's effectiveness and practicality.

加工质量和性能是数控机床的关键指标。为了提高中低端数控机床对非圆曲线和旋转零件的加工能力，提出了一种基于表面质量和几何特征约束的轮廓车削参数优化方法。该方法将轨迹优化与参数优化相结合。首先，双头蛇形（DHS）算法生成由不同曲率的直线、凸弧和凹弧组成的光滑g代码刀具路径。为提高加工质量和性能稳定性，建立了多目标优化模型。该模型采用反向传播神经网络（BPNN），基于工艺参数和曲率半径预测表面粗糙度Ra和切削力F，并分析其影响机理。提出了一种改进的非洲秃鹫多目标优化算法（MOAVOAimprove），并在CEC2009基准测试中与其他算法进行了验证，证明了其优越的性能。表面质量和几何约束指导刀具轨迹分割，将曲率对Ra影响最小的相邻轨迹段合并为单个区域。然后根据曲率和多目标解确定每个区域的优化工艺参数。将该方法应用于非圆旋转工件，并与Mastercam和基于经验的参数进行了比较。结果表明，加工质量有明显提高，验证了该方法的有效性和实用性。

{"title":"Optimization method of process parameters based on the constraints of surface quality and geometric feature of elements during copying turning","authors":"Zhiheng Chen , Ning Li , Shiying Tu , Kaimao Zeng , Juan Lu , Dan Chen","doi":"10.1016/j.asoc.2026.114711","DOIUrl":"10.1016/j.asoc.2026.114711","url":null,"abstract":"<div><div>Machining quality and performance are key indicators of CNC machine tool. To enhance the machining of non-circular curves and rotating parts on mid- to low-end CNC machines, this study proposes an optimization method for contour turning parameters based on surface quality and geometric feature constraints. The method integrates trajectory and parameter optimization. First, the Double-Heads Snake (DHS) algorithm generates a smooth G-code tool path comprising straight lines, convex arcs, and concave arcs with varying curvatures. Multi-objective optimization models are then developed for these elements to enhance machining quality and performance stability. Using Back Propagation Neural Networks (BPNN), the models predict surface roughness (<em>Ra</em>) and cutting force (<em>F</em>) based on process parameters and curvature radius, analyzing their influence mechanisms. An improved African Vulture multi-objective optimization algorithm (MOAVOA<sub>improve</sub>) is proposed and validated against other algorithms using CEC2009 benchmarks, demonstrating superior performance. Surface quality and geometric constraints guide tool path segmentation, merging adjacent path segments with minimal curvature impact on <em>Ra</em> into single regions. Optimized process parameters are then determined for each region based on curvature and multi-objective solutions. The method is applied to non-circular rotating workpieces and compared with Mastercam and experience-based parameters. Results show significant improvements in machining quality, validating the method's effectiveness and practicality.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114711"},"PeriodicalIF":6.6,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080296","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Hybrid-conditional generative adversarial network framework for climate fault detection in vertical farming environments 垂直农业环境气候故障检测的混合条件生成对抗网络框架

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-27 DOI: 10.1016/j.asoc.2026.114727

P.K.S. Tejes , Abhishek Dasore , Norhashila Hashim , Bukke Kiran Naik , Challa Monisha Reddy

Fault diagnosis in vertical farming systems (VFS) is constrained by the lack of labeled fault data, especially under rare or complex airflow disturbance conditions. This limits the training and evaluation of machine learning (ML) models for real-time microclimate monitoring and control. This study aims to develop and assess a Hybrid Conditional Generative Adversarial Network (Hybrid-CGAN) to generate synthetic microclimate sequences for specific airflow fault scenarios using computational fluid dynamics (CFD)-based simulation data. Four airflow fault scenarios, such as, reduced airflow from the upper inlet plenum, rack short-circuit, non-functional rear exhaust fan, and high static pressure at the return-duct junction, were simulated using a steady-state CFD model of a multi-tier plant factory. Diurnal perturbations generated 24-hour temporal sequences. Synthetic datasets trained Hybrid-CGAN and were compared with baselines such as Synthetic Minority Over-sampling Technique (SMOTE), Random Oversampling (ROS), and unconditional GANs. Performance was evaluated using Dynamic Time Warping (DTW), Uniform Manifold Approximation and Projection (UMAP) visualizations, and accuracy of Gated Recurrent Unit (GRU), Light Gradient Boosting Machine (LGBM), and k-Nearest Neighbors (k-NN) models. Hybrid-CGAN achieved the lowest DTW score of 0.10 at 100 % CFD data, representing a 61.03 % improvement over standard GAN (DTW = 10.45) and 59.76 % over simulation-only baseline (DTW = 10.12). Classifiers trained on Hybrid-CGAN-augmented datasets achieved accuracies of 0.95 (GRU), 0.96 (LGBM), and 0.94 (k-NN). UMAP showed high clustering similarity between real and synthetic sequences. This study establishes a practical paradigm by integrating CFD-based airflow simulations with Hybrid-CGAN framework for fault-specific synthetic data generation in VFS.

垂直农业系统（VFS）的故障诊断受到缺乏标记故障数据的限制，特别是在罕见或复杂的气流扰动条件下。这限制了用于实时微气候监测和控制的机器学习（ML）模型的训练和评估。本研究旨在开发和评估混合条件生成对抗网络（Hybrid- cgan），利用基于计算流体动力学（CFD）的模拟数据为特定的气流故障场景生成合成的微气候序列。采用多层厂房的稳态CFD模型，模拟了上进气室气流减少、机架短路、后排风机失效、回风管连接处静压过高等4种气流故障情况。日扰动产生24小时时间序列。合成数据集训练Hybrid-CGAN，并与合成少数过采样技术（SMOTE）、随机过采样（ROS）和无条件gan等基线进行比较。使用动态时间扭曲（DTW）、均匀流形逼近和投影（UMAP）可视化以及门通循环单元（GRU）、光梯度增强机（LGBM）和k-近邻（k-NN）模型的精度来评估性能。Hybrid-CGAN在100 % CFD数据下的DTW得分最低，为0.10，比标准GAN （DTW = 10.45）提高了61.03 %，比仅模拟的GAN （DTW = 10.12）提高了59.76 %。在hybrid - cgan增强数据集上训练的分类器的准确率分别为0.95 （GRU）、0.96 （LGBM）和0.94 （k-NN）。UMAP显示真实序列与合成序列具有较高的聚类相似性。该研究通过将基于cfd的气流模拟与Hybrid-CGAN框架相结合，建立了一个实用的范例，用于VFS中特定故障的合成数据生成。

{"title":"Hybrid-conditional generative adversarial network framework for climate fault detection in vertical farming environments","authors":"P.K.S. Tejes , Abhishek Dasore , Norhashila Hashim , Bukke Kiran Naik , Challa Monisha Reddy","doi":"10.1016/j.asoc.2026.114727","DOIUrl":"10.1016/j.asoc.2026.114727","url":null,"abstract":"<div><div>Fault diagnosis in vertical farming systems (VFS) is constrained by the lack of labeled fault data, especially under rare or complex airflow disturbance conditions. This limits the training and evaluation of machine learning (ML) models for real-time microclimate monitoring and control. This study aims to develop and assess a Hybrid Conditional Generative Adversarial Network (Hybrid-CGAN) to generate synthetic microclimate sequences for specific airflow fault scenarios using computational fluid dynamics (CFD)-based simulation data. Four airflow fault scenarios, such as, reduced airflow from the upper inlet plenum, rack short-circuit, non-functional rear exhaust fan, and high static pressure at the return-duct junction, were simulated using a steady-state CFD model of a multi-tier plant factory. Diurnal perturbations generated 24-hour temporal sequences. Synthetic datasets trained Hybrid-CGAN and were compared with baselines such as Synthetic Minority Over-sampling Technique (SMOTE), Random Oversampling (ROS), and unconditional GANs. Performance was evaluated using Dynamic Time Warping (DTW), Uniform Manifold Approximation and Projection (UMAP) visualizations, and accuracy of Gated Recurrent Unit (GRU), Light Gradient Boosting Machine (LGBM), and k-Nearest Neighbors (k-NN) models. Hybrid-CGAN achieved the lowest DTW score of 0.10 at 100 % CFD data, representing a 61.03 % improvement over standard GAN (DTW = 10.45) and 59.76 % over simulation-only baseline (DTW = 10.12). Classifiers trained on Hybrid-CGAN-augmented datasets achieved accuracies of 0.95 (GRU), 0.96 (LGBM), and 0.94 (k-NN). UMAP showed high clustering similarity between real and synthetic sequences. This study establishes a practical paradigm by integrating CFD-based airflow simulations with Hybrid-CGAN framework for fault-specific synthetic data generation in VFS.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114727"},"PeriodicalIF":6.6,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0