Expert Systems with Applications最新文献

{"title":"LGSMOTE-IDS: Line Graph based Weighted-Distance SMOTE for imbalanced network traffic detection","authors":"Guyu Zhao,&nbsp;Linwei Li,&nbsp;Hongdou He,&nbsp;Jiadong Ren","doi":"10.1016/j.eswa.2025.127645","DOIUrl":"10.1016/j.eswa.2025.127645","url":null,"abstract":"<div><div>The application of Graph Neural Networks (GNNs) to Network Intrusion Detection Systems (NIDS) has become a prominent research focus. However, NIDS often struggles to classify minority attack samples due to the severe class imbalance in NIDS datasets, where the number of samples varies significantly across classes. Additionally, prior studies have frequently overlooked the importance of edge features in GNNs. To address these challenges, we propose LGSMOTE-IDS, a novel framework that integrates a <strong><u>L</u></strong>ine <strong><u>G</u></strong>raph based Weighted-Distance <strong><u>SMOTE</u></strong> for <strong><u>I</u></strong>ntrusion <strong><u>D</u></strong>etection <strong><u>S</u></strong>ystems. First, we define the fine-grained protocol service graph (<span><math><mrow><mi>P</mi><mi>S</mi><mi>G</mi></mrow></math></span>) and transform it into its corresponding protocol service line graph (<span><math><mrow><mi>L</mi><mrow><mo>(</mo><mi>P</mi><mi>S</mi><mi>G</mi><mo>)</mo></mrow></mrow></math></span>). This transformation provides a novel perspective for describing network traffic interactions and enables the conversion of the edge classification task into a node classification task. Second, we introduce Weighted-Distance SMOTE, an oversampling algorithm specifically tailored to NIDS datasets, which employs an improved interpolation strategy to generate synthetic minority class samples. Finally, we utilize a GNN-based classifier to predict labels for all samples. We conduct experiments on three widely used datasets—NF-UNSW-NB15, NF-BoT-IoT, and NF-ToN-IoT. LGSMOTE-IDS achieves average increases of 18.11%, 45.91%, and 36.41% in weighted F1-scores for five, one, and three minority classes across the three datasets, respectively, compared to baseline method. Moreover, LGSMOTE-IDS successfully detects attack types that previous models fail to recognize. To the best of our knowledge, LGSMOTE-IDS is the first framework to integrate GNNs with an oversampling algorithm to address the class imbalance issue in NIDS datasets.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"281 ","pages":"Article 127645"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143838609","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}

{"title":"Accelerating Federated Learning with genetic algorithm enhancements","authors":"Huanqing Zheng ,&nbsp;Jielei Chu ,&nbsp;Zhaoyu Li ,&nbsp;Jinghao Ji ,&nbsp;Tianrui Li","doi":"10.1016/j.eswa.2025.127636","DOIUrl":"10.1016/j.eswa.2025.127636","url":null,"abstract":"<div><div>Federated Learning (FL) enables collaborative model training across multiple devices while preserving data privacy. However, developing robust and efficient FL faces significant challenges, such as data heterogeneity, computational resource constraints, communication bottlenecks, and the presence of malicious participants. To address these issues, we introduce GenFed, an innovative framework that enhances federated learning through genetic algorithm mechanisms. GenFed optimizes model aggregation strategies and balances resource utilization, thereby improving performance and resilience. This framework is designed for seamless integration with existing FL systems, facilitating rapid adaptation. GenFed accelerates model convergence and enhances robustness, particularly in environments with a large number of clients. Experimental results demonstrate that GenFed significantly outperforms traditional FL methods in terms of convergence speed, accuracy, and resilience against adversarial attacks across diverse datasets. Notably, as the number of clients increases, conventional federated methods often suffer substantial performance degradation. In contrast, GenFed maintains stable, high-level performance, making it especially practical for real-world scenarios involving extensive client participation. Our findings indicate that GenFed is a versatile and efficient solution that offers significant improvements in scalability and robustness, contributing to the deployment of reliable federated learning in real-world applications.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"281 ","pages":"Article 127636"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143838780","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}

{"title":"Micro multi-objective genetic algorithm with information fitting strategy for low-power microprocessor","authors":"Hu Peng ,&nbsp;Tian Fang ,&nbsp;Jianpeng Xiong ,&nbsp;Zhongtian Luo ,&nbsp;Tao Liu ,&nbsp;Zelin Wang","doi":"10.1016/j.eswa.2025.127644","DOIUrl":"10.1016/j.eswa.2025.127644","url":null,"abstract":"<div><div>Micro multi-objective evolutionary algorithms (<span><math><mi>μ</mi></math></span>MOEAs) are designed to address multi-objective optimization problems (MOPs), particularly in low-power microprocessor where computing resources are constrained. However, to compensate for the diversity loss resulting from using a micro population, existing optimization methods in numerous <span><math><mi>μ</mi></math></span>MOEAs lead to diminished competitiveness over time due to the absence of targeted feedback on population states, hindering further performance improvement. To address this challenge, a micro multi-objective genetic algorithm with information fitting strategy for low-power microprocessor(<span><math><mi>μ</mi></math></span>MOGAIF) is proposed, which utilizes an information fitting strategy to monitor the evolutionary status of the population and to facilitate method selection. The status information is collected at each iteration and fitted regularly, and the evaluation indicator is adjusted by the fitted evaluation results. In addition, adaptive mating selection is used in the construction of the mating pool to enhance the exploitation of solutions in probable regions. To enhance the adaptability of <span><math><mi>μ</mi></math></span>MOGAIF, dual archives are established, one archive compensates the output using various strategies to pursue convergence or diversity, while the other provides the final output set. <span><math><mi>μ</mi></math></span>MOGAIF is compared with five state-of-the-art MOEAs and five <span><math><mi>μ</mi></math></span>MOEAs on the DTLZ, WFG, MaF, and ZDT benchmark test suites, and the experimental results demonstrate that <span><math><mi>μ</mi></math></span>MOGAIF has outstanding performance. Furthermore, simulations based on low-power microprocessor have been conducted to verify the feasibility of <span><math><mi>μ</mi></math></span>MOGAIF.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127644"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864689","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}

{"title":"Nonnegative graph embedding induced unsupervised feature selection","authors":"Yong Mi ,&nbsp;Hongmei Chen ,&nbsp;Zhong Yuan ,&nbsp;Chuan Luo ,&nbsp;Shi-Jinn Horng ,&nbsp;Tianrui Li","doi":"10.1016/j.eswa.2025.127664","DOIUrl":"10.1016/j.eswa.2025.127664","url":null,"abstract":"<div><div>Recently, many unsupervised feature selection (UFS) methods have been developed due to their effectiveness in selecting valuable features to improve and accelerate the subsequent learning tasks. However, most existing UFS methods suffer from the following three drawbacks: (1) They usually ignore the nonnegative attribute of feature when conducting feature selection, which inevitably loses partial information; (2) Most adopt a separate strategy to rank all features and then select the first <span><math><mi>k</mi></math></span> features, which introduces an additional parameter and often obtains suboptimal results; (3) Most generally confront the problem of high time-consuming. To tackle the previously mentioned shortage, we present a novel UFS method, <em>i</em>.<em>e</em>., Nonnegative Graph Embedding Induced Unsupervised Feature Selection, which considers nonnegative feature attributes and selects informative feature subsets in a one-step way. Specifically, the raw data are projected into a low-dimensional subspace, where the learned low-dimensional representation keeps a nonnegative attribute. Then, a novel scheme is designed to preserve the local geometric structure of the original data, and <span><math><msub><mrow><mi>ℓ</mi></mrow><mrow><mn>2</mn><mo>,</mo><mn>0</mn></mrow></msub></math></span> norm is introduced to guide feature selection without ranking and selecting processes. Finally, we design a high-efficiency solution strategy with low computational complexity, and experiments on real-life datasets verify the efficiency and advancement compared with advanced UFS methods.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127664"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143869644","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}

{"title":"Facial expression recognition based on multi-task self-distillation with coarse and fine grained labels","authors":"Ziyang Zhang ,&nbsp;Xu Li ,&nbsp;Kailing Guo ,&nbsp;Xiangmin Xu","doi":"10.1016/j.eswa.2025.127440","DOIUrl":"10.1016/j.eswa.2025.127440","url":null,"abstract":"<div><div>Facial expression recognition (FER) plays a crucial role in numerous human–computer interaction systems. For the sake of precise recognition, existing methods often enhance the representational capacity of networks by designing complex network structures or incorporating additional facial information. However, due to redundancy among facial expression features, refining expression-related information to obtain highly discriminative expression features remains challenging. We propose a multi-task self-distillation method with coarse and fine grained labels for FER. To mine the sufficient expression-related information, we construct coarse-grained auxiliary branches that enhance the learning ability of the network based on the prior in the facial expression labels. To map coarse-grained features into a fine-grained feature space, feature alignment modules are then introduced. Then, refined self-distillation is constructed to transfer coarse-grained knowledge to fine-grained features, providing additional guidance for the extraction of discriminative features. Our proposed method achieves state-of-the-art performance on multiple FER benchmarks, demonstrating its superiority.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"281 ","pages":"Article 127440"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143842730","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}

{"title":"IIBNet: Inter- and Intra-Information balance network with Self-Knowledge distillation and region attention for RGB-D indoor scene parsing","authors":"Yuming Zhang ,&nbsp;Fangfang Qiang ,&nbsp;Wujie Zhou ,&nbsp;Weiqing Yan ,&nbsp;Lv Ye","doi":"10.1016/j.eswa.2025.127670","DOIUrl":"10.1016/j.eswa.2025.127670","url":null,"abstract":"<div><div>Red–green–blue-depth (RGB-D) indoor scene parsing is a vital research topic in computer vision. Here, features acquired from different layers of the backbone are further processed to obtain a better prediction image. These images have different sizes and contain different information, but the labels used to supervise them are the same, resulting in large inconsistencies. Moreover, in training images, several categories are represented by a relatively small proportion of the total number of categories, resulting in poor training results. To address these problems, this article proposes an inter- and intra-information balance network (IIBNet). First, to offset the category imbalance within features, a region balance module using a region attention module is employed to merge four pairs of features obtained from the RGB and depth backbone networks, adjusting the proportion of different categories allocated in the feature. Second, to address the problem of feature-information imbalance across layers, information is transferred between two branches, reducing the diversity of information across different layers. The first branch is a channel-wise information-interaction branch, which employs self-knowledge distillation (Self-KD) as a tool for information transfer. Self-KD, in which the student and teacher networks are the same, allows features to learn from each other. The second branch is a spatial-wise information-interaction branch, which transfers the lowest-level feature information to the higher-level features. Based on extensive testing on two large indoor-scene-parsing datasets, IIBNet is observed to outperform state-of-the-art methods on three metrics.<!--> <!-->The source code and results are available at <span><span>https://github.com/kolaloaver/IIBNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127670"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143851907","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}

{"title":"Multimodal speech emotion recognition via dynamic multilevel contrastive loss under local enhancement network","authors":"Weiquan Fan ,&nbsp;Xiangmin Xu ,&nbsp;Fang Liu ,&nbsp;Xiaofen Xing","doi":"10.1016/j.eswa.2025.127669","DOIUrl":"10.1016/j.eswa.2025.127669","url":null,"abstract":"<div><div>Multimodal speech emotion recognition is crucial for advancing human–computer interaction technology. Contrastive learning, due to its powerful ability of representation, is increasingly being applied to emotion recognition. Existing algorithms usually only consider samples of the same emotion as positive matching pairs, but ignore that the distances of different positive pairs are often different. For this issue, this paper designs a novel dynamic multilevel contrastive loss (DMCL), which achieves adaptive distance constraint by dynamic multilevel similarity. It generalizes positive matching pairs in different cases, assigns them different distances, and dynamically adjusts the corresponding labels while modeling. Building upon the DMCL, this paper further proposes a local enhancement attention mechanism (LEA) that enhances local information token-by-token on a global basis, which can enhance the robustness of the model to emotional mutations. By integrating the advantages of LEA and DMCL, this paper constructs an end-to-end multimodal speech emotion recognition network (LEDMCN). Finally, experimental results on the IEMOCAP and LSSED datasets validate the effectiveness of the proposed method, achieving state-of-the-art performance.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"281 ","pages":"Article 127669"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143848626","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}

{"title":"Flexible anchor-based trajectory prediction for different types of traffic participants in autonomous driving systems","authors":"Yingjuan Tang ,&nbsp;Hongwen He ,&nbsp;Yong Wang ,&nbsp;Yifan Wu","doi":"10.1016/j.eswa.2025.127629","DOIUrl":"10.1016/j.eswa.2025.127629","url":null,"abstract":"<div><div>The task of trajectory prediction is a critical component of autonomous vehicle systems. Existing trajectory prediction methodologies encounter challenges in effectively handling varied traffic participant categories and accurately forecasting long-term trajectories. In response, we introduce the Fourier Transformer Prediction (FTP) framework, which integrates the Fourier transform and a flexible anchor approach. The Fourier transform encoder adeptly captures temporal and spectral domain features inherent in trajectory data across diverse categories. The flexible anchor method employs a proposal module without anchors to generate adaptable coarse trajectories, complemented by an anchor-based module for subsequent refinement. FTP adeptly models the characteristics of multi-class participants, enhancing training stability and mitigating issues such as mode collapse. Through extensive experiments conducted on public datasets and the proposed Traffic Route Bus trajectory prediction dataset (TRB), FTP demonstrates superior performance, underscoring its efficacy across diverse traffic scenarios.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127629"},"PeriodicalIF":7.5,"publicationDate":"2025-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143859189","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}

{"title":"ResDNViT: A hybrid architecture for Netflow-based attack detection using a residual dense network and Vision Transformer","authors":"Hassan Wasswa,&nbsp;Hussein A. Abbass,&nbsp;Timothy Lynar","doi":"10.1016/j.eswa.2025.127504","DOIUrl":"10.1016/j.eswa.2025.127504","url":null,"abstract":"<div><div>The fast evolution of technologies like wireless sensor networks, cloud computing services, advanced AI driven applications and the Internet of Things (IoT) have led to increased reliance on internet by both individual users and enterprises—both small and large. On the contrary, the advancements in cybersecurity have not matched this pace consequently attracting exponentially rising trends of cyberattacks in the past decade. To enhance network security, this work proposes ResDNViT, a robust model integrating a self-attention-based Vision Transformer (ViT) architecture with a simplified ResNet-based architecture for NetFlow-based attack detection. Motivated by the strong performance of transformers in tasks related to NLP and computer vision, ResDNViT extends the ViT-based architecture for network traffic analysis by expressing NetFlow features as 2D matrices, and splitting them into equal-sized sub-matrices, that are used as input patches for the encoder component. A simplified residual dense network (ResDN) with two residual dense blocks (RDB) is stacked to the encoder’s output layer for classification. The novelty of this approach lies in effectively adapting the ViT-based architecture, originally designed for images, to analyzing NetFlow packets for attack classification. The model was evaluated on four well-studied benchmark datasets: the CICIDS2017_improved, Bot-IoT, CICIoT2022, and N-BaIoT, demonstrating an impressive performance across various classification tasks. The proposed approach’s ability to detect traffic from unseen device kinds was assessed by grouping devices from N-BaIoT into five categories based on usage: Thermostats, Baby Monitors, Doorbells, Security Cameras and Webcams. The model was trained using samples from four categories at a time and tested on samples from the remaining category. A high performance across metrics including accuracy, precision, recall, and F1-score for all categories highlighted the model’s robustness in traffic discrimination.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127504"},"PeriodicalIF":7.5,"publicationDate":"2025-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143851830","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}

{"title":"A global and local unified feature selection algorithm based on hierarchical structure constraints","authors":"Yibin Wang ,&nbsp;Xinru Zhang ,&nbsp;Yusheng Cheng","doi":"10.1016/j.eswa.2025.127535","DOIUrl":"10.1016/j.eswa.2025.127535","url":null,"abstract":"<div><div>Existing feature selection methods face challenges when applied to hierarchically structured data, which can be primarily due to a lack of synergy between nodes, resulting in impaired global consistency and poor local coherence. For example, parent nodes may dominate feature weighting without child feedback (e.g., suppressing texture details in fine-grained image classification), while sibling nodes fail to capture asymmetric dependencies in shared features (e.g., genetic markers varying across disease subtypes). To address these issues, a Global and Local Unified Feature Selection algorithm was proposed based on Hierarchical Structure Constraints (GLUFS-HSC). This algorithm integrated global and local perspectives and introduced a bidirectional consistency constraint mechanism for parent–child nodes, along with an asymmetry constraint mechanism for sibling nodes. These innovations enhanced feature selection efficiency and inter-level coordination. The algorithm employed a multi-objective optimization framework to maintain consistency while preserving the original data features. At the global level, it incorporated node relationships and hierarchical requirements by iteratively updating a weight matrix. At the local level, the traditional one-way dependency or implicit bidirectional models were replaced with an explicit parent–child bidirectional consistency constraint, enabling the parent nodes to dynamically adjust the weight distribution based on feedback from child nodes. This approach facilitated information transfer and strengthened hierarchical synergy. For sibling nodes, an asymmetric constraint mechanism combining HSIC constraint and orthogonal constraint is introduced to effectively capture feature differences, reduce feature redundancy, and enhance feature independence and correlation. Experimental comparisons across eight datasets demonstrated that GLUFS-HSC achieved superior performance on hierarchically structured data, significantly improving the consistency and accuracy of feature selection.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127535"},"PeriodicalIF":7.5,"publicationDate":"2025-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143851817","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}