Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 260 - 275
Опубликована: Окт. 17, 2024
Язык: Английский
Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 260 - 275
Опубликована: Окт. 17, 2024
Язык: Английский
Cluster Computing, Год журнала: 2025, Номер 28(4)
Опубликована: Фев. 25, 2025
Язык: Английский
Процитировано
2Sensors, Год журнала: 2025, Номер 25(3), С. 854 - 854
Опубликована: Янв. 30, 2025
Wireless communication advancements have significantly improved connectivity and user experience with each generation. The recent release of the framework M.2160 for upcoming sixth generation (6G or IMT-2030) cellular wireless standard by ITU-R has heightened expectations, particularly Internet Things (IoT) driven use cases. However, this progress introduces significant security risks, as technologies like O-RAN, terahertz communication, native AI pose threats such eavesdropping, supply chain vulnerabilities, model poisoning, adversarial attacks. increased exposure sensitive data in 6G applications further intensifies these challenges. This necessitates a concerted effort from stakeholders including ITU-R, 3GPP, ETSI, OEMs researchers to embed resilience core components 6G. While research is advancing, establishing comprehensive remains challenge. To address evolving threats, our proposes dynamic that emphasizes integration explainable (XAI) techniques SHAP LIME advanced machine learning models enhance decision-making transparency, improve complex environments, ensure effective detection mitigation emerging cyber threats. By refining accuracy ensuring alignment through recursive feature elimination consistent cross-validation, approach strengthens overall posture IoT–6G ecosystem, making it more resilient attacks other vulnerabilities.
Язык: Английский
Процитировано
1Sensors, Год журнала: 2025, Номер 25(4), С. 1039 - 1039
Опубликована: Фев. 9, 2025
Internet of Things (IoT) networks’ wide range and heterogeneity make them prone to cyberattacks. Most IoT devices have limited resource capabilities (e.g., memory capacity, processing power, energy consumption) function as conventional intrusion detection systems (IDSs). Researchers applied many approaches lightweight IDSs, including energy-based machine learning/deep learning (ML/DL)-based federated (FL)-based IDSs. FL has become a promising solution for IDSs in networks because it reduces the overhead process by engaging during training process. Three architectures are used tackle networks, centralized (client–server), decentralized (device-to-device), semi-decentralized. However, none solved while considering lightweight-ness performance at same time. Therefore, we propose semi-decentralized FL-based model IDS fit device capabilities. The proposed is based on clustering devices—FL clients—and assigning cluster head each that acts behalf clients. Consequently, number communicate with server reduced, helping reduce communication overhead. Moreover, helps improving aggregation sends average model’s weights one round. distributed denial-of-service (DDoS) attack main concern our model, since easily occurs configured three deep techniques—LSTM, BiLSTM, WGAN—using CICIoT2023 dataset. experimental results show BiLSTM achieves better suitable resource-constrained size. We test pre-trained datasets—BoT-IoT, WUSTL-IIoT-2021, Edge-IIoTset—and highest most classes, particularly DDoS attacks.
Язык: Английский
Процитировано
1Informatics, Год журнала: 2025, Номер 12(1), С. 17 - 17
Опубликована: Фев. 11, 2025
This paper presents an intrusion detection system (IDS) leveraging a hybrid machine learning approach aimed at enhancing the security of IoT devices edge, specifically for those utilizing TCP/IP protocol. Recognizing critical challenges posed by rapid expansion networks, this work evaluates proposed IDS model with primary focus on optimizing training time without sacrificing accuracy. The begins comprehensive review existing models IDS, highlighting both their strengths and limitations. It then provides overview technologies methodologies implemented in work, including utilization “Botnet Traffic Dataset For Smart Buildings”, newly released public dataset tailored threat detection. is explained detail, followed discussion experimental results that assess model’s performance real-world conditions. Furthermore, evaluated its effectiveness within smart building environments, demonstrating how it can address unique such as resource constraints real-time edge. aims to contribute development efficient, reliable, scalable solutions protect ecosystems from emerging threats.
Язык: Английский
Процитировано
0Technologies, Год журнала: 2025, Номер 13(4), С. 141 - 141
Опубликована: Апрель 4, 2025
Bank fraud detection faces critical challenges in imbalanced datasets, where fraudulent transactions are rare, severely impairing model generalization. This study proposes a Gaussian noise-based augmentation method to address class imbalance, contrasting it with SMOTE and ADASYN. By injecting controlled perturbations into the minority class, our approach mitigates overfitting risks inherent interpolation-based techniques. Five classifiers, including XGBoost convolutional neural network (CNN), were evaluated on augmented datasets. achieved superior performance noise-augmented data (accuracy: 0.999507, AUC: 0.999506), outperforming These results underscore noise’s efficacy enhancing accuracy, offering robust alternative conventional oversampling methods. Our findings emphasize pivotal role of strategies optimizing classifier for financial data.
Язык: Английский
Процитировано
0Electronics, Год журнала: 2025, Номер 14(7), С. 1459 - 1459
Опубликована: Апрель 4, 2025
With the increase in ownership of Internet Things (IoT) devices, there is a bigger demand for stronger implementation security mechanisms and addressing zero-day vulnerabilities. This work first to provide platform that combines featureless approaches with artificial intelligence (AI) algorithms, which are deep learning large language models, uncover IoT vulnerabilities based on network traffic data directly without manual feature selection. The correctly identifies vulnerable secure devices just by raw traffic! Experimental results show proposed study detects vulnerability great accuracy using pre-trained LLM facilitates direct extraction features from dataset therefore helps speed up identification process. In addition, design ensures models accessible can be easily applied users user-friendly interface. Furthermore, small sizes, 277.5 MB 334 model model, respectively, illustrated potential use detection tool practical settings. ability defend large-scale, diversified ecosystems efficiently scalable way installing thousands software manifestations quickly while exposing new applications growing cyber threats made possible this significant advancement field security.
Язык: Английский
Процитировано
0Future Internet, Год журнала: 2025, Номер 17(6), С. 234 - 234
Опубликована: Май 26, 2025
An intrusion detection system (IDS) is a crucial element in cyber security concerns. IDS safeguarding module that designed to identify unauthorized activities network environments. The importance of constructing IDSs has never been this significant with the growing number attacks on layers. This research work was intended draw attention authors different aspect detection, considering privacy and contribution features attack classes. At present, majority existing are based centralized infrastructure, which raises serious concerns about as data from one exposed another system. act sharing original server can worsen current arrangement protecting within network. In addition, models merely tool for identifying categories without analyzing further emphasis feature attacks. article, we propose novel framework, FEDXAIIDS, converging federated learning explainable AI. proposed approach enables be collaboratively trained across multiple decentralized devices while ensuring local remain securely edge nodes, thus mitigating risks. primary objectives study reveal systems most comprehend final output. Our model designed, fusing (FL) Shapley additive explanations (SHAPs), using an artificial neural (ANN) model. framework device four client have their own set end. distributes constructed ANN among clients. Next, clients train individual part set, deploying distributed server, they share feedback central end then incorporates aggregator named FedAvg assemble separate results into last, ten evaluated by incorporating SHAP. entire executed CICIoT2023. partitioned parts ends. method demonstrated efficacy achieving 88.4% training 88.2% testing accuracy. Furthermore, UDP found layer SHAP analysis. Simultaneously, incorporation ensured confidentiality information enhances transparency ensures both reliable interpretable. Federated XAI effectively addresses interpretability issues modern frameworks, contributing advancement secure, interpretable, systems. findings accelerate development solutions leverage AI (XAI), paving way future practical implementations real-world
Язык: Английский
Процитировано
0Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 260 - 275
Опубликована: Окт. 17, 2024
Язык: Английский
Процитировано
1