Pervasive and Mobile Computing最新文献_第4页

SmartCert: A Multi-modal framework for automated guided vehicle screening SmartCert：用于自动引导车辆筛选的多模式框架

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-28 DOI: 10.1016/j.pmcj.2025.102127

Xu Chen , Sandeep Kanta , Vincent Koc , Santhi Bharath Punati , Arif Hussain , Sunny Katyara

Global used vehicle market is undergoing rapid transformation with proliferating demand for scalable, efficient and trustworthy inspection systems that is capable of meeting stringent requirements of online marketplaces and regulatory standards. This paper introduces SmartCert, a multi-modal inspection framework engineered for robust, scalable and pervasive vehicle screening. The novelty of SmartCert lies in synergistic integration of tailored multi-modal transformer architecture with fine-grained temporal diagnostics and optimized edge deployment. An embedded cross-attention mechanism fosters seamless fusion of visual data with on-board diagnostic signals to simultaneously detect exterior damages and internal performance anomalies. To ensure reliable evaluation, SmartCert incorporates reinforcement learning agent with human-in-the-loop reward scheme for adaptive certification thresholding that reduces false positive rates by 6.8% and false negative rates by 4.2% compared to optimally tuned static thresholds. Rigorously evaluated on large-scale dataset of 10240 vehicles with edge deployment validated exclusively on 240 vehicles (2.3%) collected from diverse mobile inspection locations, SmartCert achieves F1-score of 95% for damage classification and 92% anomaly detection rate. These results demonstrate statistically significant improvements over same-dataset baseline implementations by average of 7.4% in classification accuracy and 9.6% in anomaly detection (

p < 0.001

). Furthermore in ablation study, SmartCert improves processing efficiency by 40%, reduces certification-to-sale by 30% and decreases post-sale complaints by 25% compared to traditional manual methods. By integrating explainable AI with optimized edge deployment achieve 18 FPS inference on resource-constrained hardware, SmartCert articulates end-to-end solution for next generation of trustworthy and efficient vehicle certification ecosystem.

全球二手车市场正在经历快速转型，对可扩展、高效和值得信赖的检测系统的需求激增，这些系统能够满足在线市场和监管标准的严格要求。本文介绍了SmartCert，这是一种多模态检查框架，用于强大，可扩展和普遍的车辆筛选。SmartCert的新颖之处在于将定制的多模态变压器架构与细粒度的时间诊断和优化的边缘部署协同集成。嵌入式交叉关注机制促进视觉数据与车载诊断信号的无缝融合，同时检测外部损伤和内部性能异常。为了确保可靠的评估，SmartCert将强化学习代理与人在环奖励方案结合起来，用于自适应认证阈值，与优化的静态阈值相比，可将假阳性率降低6.8%，假阴性率降低4.2%。在10240辆车的大规模数据集上进行严格评估，仅对来自不同移动检测地点的240辆车（2.3%）进行边缘部署验证，SmartCert的损伤分类得分为f1 - 95%，异常检测率为92%。这些结果表明，与相同数据集基线实现相比，分类准确率平均提高了7.4%，异常检测平均提高了9.6% （p<0.001）。此外，在消融研究中，与传统的手工方法相比，SmartCert将处理效率提高了40%，将认证到销售的时间缩短了30%，将售后投诉减少了25%。通过集成可解释的人工智能和优化的边缘部署，在资源受限的硬件上实现18 FPS推理，SmartCert为下一代值得信赖和高效的车辆认证生态系统提供了端到端解决方案。

{"title":"SmartCert: A Multi-modal framework for automated guided vehicle screening","authors":"Xu Chen , Sandeep Kanta , Vincent Koc , Santhi Bharath Punati , Arif Hussain , Sunny Katyara","doi":"10.1016/j.pmcj.2025.102127","DOIUrl":"10.1016/j.pmcj.2025.102127","url":null,"abstract":"<div><div>Global used vehicle market is undergoing rapid transformation with proliferating demand for scalable, efficient and trustworthy inspection systems that is capable of meeting stringent requirements of online marketplaces and regulatory standards. This paper introduces SmartCert, a multi-modal inspection framework engineered for robust, scalable and pervasive vehicle screening. The novelty of SmartCert lies in synergistic integration of tailored multi-modal transformer architecture with fine-grained temporal diagnostics and optimized edge deployment. An embedded cross-attention mechanism fosters seamless fusion of visual data with on-board diagnostic signals to simultaneously detect exterior damages and internal performance anomalies. To ensure reliable evaluation, SmartCert incorporates reinforcement learning agent with human-in-the-loop reward scheme for adaptive certification thresholding that reduces false positive rates by 6.8% and false negative rates by 4.2% compared to optimally tuned static thresholds. Rigorously evaluated on large-scale dataset of 10240 vehicles with edge deployment validated exclusively on 240 vehicles (2.3%) collected from diverse mobile inspection locations, SmartCert achieves F1-score of 95% for damage classification and 92% anomaly detection rate. These results demonstrate statistically significant improvements over same-dataset baseline implementations by average of 7.4% in classification accuracy and 9.6% in anomaly detection (<span><math><mrow><mi>p</mi><mo><</mo><mn>0</mn><mo>.</mo><mn>001</mn></mrow></math></span>). Furthermore in ablation study, SmartCert improves processing efficiency by 40%, reduces certification-to-sale by 30% and decreases post-sale complaints by 25% compared to traditional manual methods. By integrating explainable AI with optimized edge deployment achieve 18 FPS inference on resource-constrained hardware, SmartCert articulates end-to-end solution for next generation of trustworthy and efficient vehicle certification ecosystem.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"115 ","pages":"Article 102127"},"PeriodicalIF":3.5,"publicationDate":"2025-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145468155","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

DeSIST: Emergent security in IoT through Decentralized Strategic Interactions — A game-theoretic Zero Trust framework DeSIST：通过分散战略交互的物联网应急安全——博弈论零信任框架

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-28 DOI: 10.1016/j.pmcj.2025.102124

Seyed Hossein Ahmadpanah , Meghdad Mirabi , Sanaz Sobhanloo , Pania Afsharfarnia , Donya Fallah

Numerous Internet of Things (IoT) devices are connecting our world, but they also introduce new security risks—particularly in large, power-constrained networks where traditional security techniques often fail. This paper proposes a new framework for IoT security, called DeSIST (Decentralized Strategic Interaction for Secure IoT), which is grounded in game theory and Zero Trust principles. Unlike approaches that rely on centralized watchdogs or explicit trust scores, DeSIST models interactions between IoT nodes as a sequence of strategic games. Treating nodes as rational agents, each maximizes its own expected utility when making decisions. Security emerges naturally because these games are designed with reward systems that encourage cooperation among trustworthy nodes while strategically isolating malicious or non-compliant actors. DeSIST employs a lightweight Local Information Assessor (LIA) to collect immediate, local, and contextually relevant information about ongoing or anticipated interactions, and a Strategic Decision Unit (SDU) to evaluate possible strategies and select the one that maximizes expected utility. Through theoretical analysis and extensive simulations, we show that DeSIST can decentralize and resource-efficiently uphold Zero Trust principles while significantly enhancing network resilience against common IoT attacks. Compared to existing approaches, the simulation results demonstrate notable improvements in both security and performance across different attack scenarios. DeSIST provides a promising path toward strong, incentive-driven, and emergent security in the evolving IoT landscape.

许多物联网（IoT）设备正在连接我们的世界，但它们也带来了新的安全风险——特别是在传统安全技术经常失效的大型、功率受限的网络中。本文提出了一个基于博弈论和零信任原则的物联网安全新框架，称为DeSIST （Decentralized Strategic Interaction for Secure IoT）。与依赖集中监管机构或明确信任评分的方法不同，DeSIST将物联网节点之间的交互建模为一系列战略博弈。将节点视为理性代理，每个节点在做出决策时都最大化自己的预期效用。安全性自然出现，因为这些游戏设计了奖励系统，鼓励可信节点之间的合作，同时战略性地隔离恶意或不合规的参与者。DeSIST使用轻量级的本地信息评估器（Local Information Assessor， LIA）来收集有关正在进行的或预期的交互的即时、本地和上下文相关的信息，并使用战略决策单元（Strategic Decision Unit， SDU）来评估可能的策略并选择最大化预期效用的策略。通过理论分析和广泛的模拟，我们表明DeSIST可以去中心化和资源高效地维护零信任原则，同时显着增强网络抵御常见物联网攻击的弹性。与现有方法相比，仿真结果表明该方法在不同攻击场景下的安全性和性能都有显著提高。在不断发展的物联网环境中，DeSIST为实现强大、激励驱动和紧急安全提供了一条有希望的道路。

{"title":"DeSIST: Emergent security in IoT through Decentralized Strategic Interactions — A game-theoretic Zero Trust framework","authors":"Seyed Hossein Ahmadpanah , Meghdad Mirabi , Sanaz Sobhanloo , Pania Afsharfarnia , Donya Fallah","doi":"10.1016/j.pmcj.2025.102124","DOIUrl":"10.1016/j.pmcj.2025.102124","url":null,"abstract":"<div><div>Numerous Internet of Things (IoT) devices are connecting our world, but they also introduce new security risks—particularly in large, power-constrained networks where traditional security techniques often fail. This paper proposes a new framework for IoT security, called DeSIST (Decentralized Strategic Interaction for Secure IoT), which is grounded in game theory and Zero Trust principles. Unlike approaches that rely on centralized watchdogs or explicit trust scores, DeSIST models interactions between IoT nodes as a sequence of strategic games. Treating nodes as rational agents, each maximizes its own expected utility when making decisions. Security emerges naturally because these games are designed with reward systems that encourage cooperation among trustworthy nodes while strategically isolating malicious or non-compliant actors. DeSIST employs a lightweight Local Information Assessor (LIA) to collect immediate, local, and contextually relevant information about ongoing or anticipated interactions, and a Strategic Decision Unit (SDU) to evaluate possible strategies and select the one that maximizes expected utility. Through theoretical analysis and extensive simulations, we show that DeSIST can decentralize and resource-efficiently uphold Zero Trust principles while significantly enhancing network resilience against common IoT attacks. Compared to existing approaches, the simulation results demonstrate notable improvements in both security and performance across different attack scenarios. DeSIST provides a promising path toward strong, incentive-driven, and emergent security in the evolving IoT landscape.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"115 ","pages":"Article 102124"},"PeriodicalIF":3.5,"publicationDate":"2025-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145420338","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

XAI-driven multi-attention DeepCRNN for enhanced cyberattack detection in internet of medical things environments xai驱动的多关注深度神经网络用于增强医疗物联网环境下的网络攻击检测

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-16 DOI: 10.1016/j.pmcj.2025.102123

Prashant Giridhar Shambharkar , Nikhil Sharma

The rapid proliferation of Internet of Medical Things (IoMT) devices has transformed healthcare by enabling continuous monitoring and intelligent data exchange, but it has also broadened the attack surface for cyber intrusions. Conventional intrusion detection systems (IDS) face critical challenges such as high-dimensional and imbalanced traffic patterns, dynamic data distributions, and limited adaptability in real-world IoMT settings. To overcome these limitations, we propose MA-DeepCRNN, a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs), Bidirectional LSTMs (Bi-LSTMs), and a multi-attention mechanism for robust binary and multiclass intrusion detection. The model employs a four-stage preprocessing pipeline incorporating feature augmentation, Gaussian noise injection, and categorical randomization to improve data balance and resilience. The Performance is further enhanced through epoch and batch size tuning, while an ablation study and statistical significance tests validate architectural effectiveness. Moreover, computational complexity analysis ensures suitability for resource-constrained IoMT environments, and a dual-layer explainable AI approach offers interpretability for security analysts. The Extensive experiments on the WUSTL-HDRL-2024 dataset demonstrate superior outcomes, achieving 0.9979 accuracy and 0.9966 F1-score in binary classification and 0.9823 accuracy and 0.9812 F1-score in multiclass detection. Compared with state-of-the-art, MA-DeepCRNN delivers 6–12% higher accuracy and 7–10% higher F1-score, with an overall improvement of 6.17% accuracy and 7.91% F1-score. These results establish MA-DeepCRNN as a statistically validated, interpretable, and computationally efficient IDS for real-time IoMT cybersecurity.

医疗物联网（IoMT）设备的快速普及通过实现持续监控和智能数据交换，改变了医疗保健行业，但它也扩大了网络入侵的攻击面。传统的入侵检测系统（IDS）面临着严峻的挑战，如高维和不平衡的流量模式、动态数据分布以及在实际IoMT环境中的有限适应性。为了克服这些限制，我们提出了MA-DeepCRNN，这是一种混合深度学习框架，它集成了卷积神经网络（cnn）、双向LSTMs （Bi-LSTMs）和多注意机制，用于鲁棒的二进制和多类入侵检测。该模型采用四阶段预处理流程，包括特征增强、高斯噪声注入和分类随机化，以改善数据平衡和弹性。性能通过epoch和批大小调优得到进一步增强，同时消融研究和统计显著性测试验证了体系结构的有效性。此外，计算复杂性分析确保了资源受限的IoMT环境的适用性，双层可解释的AI方法为安全分析师提供了可解释性。在WUSTL-HDRL-2024数据集上进行的大量实验显示了较好的结果，二分类准确率为0.9979，f1分数为0.9966；多类检测准确率为0.9823，f1分数为0.9812。与最先进的技术相比，MA-DeepCRNN的准确率提高了6-12%，f1分数提高了7-10%，总体准确率提高了6.17%，f1分数提高了7.91%。这些结果表明，MA-DeepCRNN是一种经过统计验证的、可解释的、计算效率高的实时IoMT网络安全IDS。

{"title":"XAI-driven multi-attention DeepCRNN for enhanced cyberattack detection in internet of medical things environments","authors":"Prashant Giridhar Shambharkar , Nikhil Sharma","doi":"10.1016/j.pmcj.2025.102123","DOIUrl":"10.1016/j.pmcj.2025.102123","url":null,"abstract":"<div><div>The rapid proliferation of Internet of Medical Things (IoMT) devices has transformed healthcare by enabling continuous monitoring and intelligent data exchange, but it has also broadened the attack surface for cyber intrusions. Conventional intrusion detection systems (IDS) face critical challenges such as high-dimensional and imbalanced traffic patterns, dynamic data distributions, and limited adaptability in real-world IoMT settings. To overcome these limitations, we propose MA-DeepCRNN, a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs), Bidirectional LSTMs (Bi-LSTMs), and a multi-attention mechanism for robust binary and multiclass intrusion detection. The model employs a four-stage preprocessing pipeline incorporating feature augmentation, Gaussian noise injection, and categorical randomization to improve data balance and resilience. The Performance is further enhanced through epoch and batch size tuning, while an ablation study and statistical significance tests validate architectural effectiveness. Moreover, computational complexity analysis ensures suitability for resource-constrained IoMT environments, and a dual-layer explainable AI approach offers interpretability for security analysts. The Extensive experiments on the WUSTL-HDRL-2024 dataset demonstrate superior outcomes, achieving 0.9979 accuracy and 0.9966 F1-score in binary classification and 0.9823 accuracy and 0.9812 F1-score in multiclass detection. Compared with state-of-the-art, MA-DeepCRNN delivers 6–12% higher accuracy and 7–10% higher F1-score, with an overall improvement of 6.17% accuracy and 7.91% F1-score. These results establish MA-DeepCRNN as a statistically validated, interpretable, and computationally efficient IDS for real-time IoMT cybersecurity.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102123"},"PeriodicalIF":3.5,"publicationDate":"2025-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145362446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Asymptotically efficient ADMM solutions for source localization using RSS measurements 使用RSS测量进行源定位的渐近有效ADMM解决方案

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-14 DOI: 10.1016/j.pmcj.2025.102122

Xiaoping Wu , Xiang Wang , Lingfang Kong, Keqi Zhou

Received Signal Strength (RSS) measurements are widely applied in wireless localization. In this paper, standard form of Alternating Direction Method of Multipliers (ADMM) is designed for source localization using RSS. The Maximum Likelihood (ML) estimation problem of RSS-based localization is equivalent to the standard ADMM form by defining the intermediate variables. Following this, we develop the solutions to the subproblems in the ADMM structure. The convergence of the proposed ADMM solution is discussed based on the convexity analysis of the subproblems, providing the evidence for its stable performance. The simulated results show that the ADMM solution performs efficiently, especially with a small number of sensors or in the presence of high noise levels. In addition, we also verify the bias performance in the source position estimation.

接收信号强度（RSS）测量在无线定位中有着广泛的应用。本文设计了一种标准形式的交替方向乘法器（ADMM），用于RSS源定位。通过定义中间变量，将基于rss的定位的最大似然估计问题等效为标准的ADMM形式。在此基础上，给出了ADMM结构中子问题的求解方法。基于子问题的凸性分析，讨论了所提ADMM解的收敛性，为其稳定性提供了证据。仿真结果表明，在传感器数量较少或存在高噪声水平的情况下，ADMM方案具有良好的性能。此外，我们还验证了在源位置估计中的偏差性能。

引用次数: 0

WiKAN: Lightweight Kolmogorov–Arnold Networks for accurate indoor WiFi localization WiKAN：轻量级Kolmogorov-Arnold网络，用于精确的室内WiFi定位

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-04 DOI: 10.1016/j.pmcj.2025.102121

Yunlong Gu , Meng Xu , Jiguang Li , Qilei Li , Zhao Huang , Mengshan Li , Lixin Guan , Mikko Valkama

With the growing demand for location-based services, WiFi localization plays a critical role in indoor environments. While most existing methods rely on Multi-Layer Perceptrons (MLPs), these models often suffer from limited accuracy and poor generalization across diverse deployment conditions. Kolmogorov–Arnold Networks (KANs), with their B-spline-based basis functions, better capture complex nonlinear relationships while reducing overfitting risks. However, original KANs still incur high computational costs. To address this, we propose WiKAN(WiFi KAN), a lightweight KAN-based model for indoor WiFi localization. WiKAN reduces computational complexity by simplifying the network structure to just two KANLinear layers and replacing parameter-intensive operations with optimized matrix multiplications using reconstructed basis functions. Compared to conventional computation of basis coefficients, matrix operations enable faster inference on modern hardware and improve scalability. Furthermore, WiKAN integrates SiLU and B-spline activations through a learnable linear combination, balancing smooth approximation and nonlinear representation. Experiments on three benchmark datasets (UJIIndoorLoc, Tampere, and JARIL) demonstrate that WiKAN achieves superior performance to both MLP and standard KAN models: over 99.9% building accuracy, up to 100% floor classification, and average positioning error reduced to 5.91 meters. Additionally, runtime analysis and parameter count comparisons confirm the model’s computational efficiency. Code is publicly available at: https://github.com/gyl555666/WiKAN.

随着定位服务需求的不断增长，WiFi定位在室内环境中发挥着至关重要的作用。虽然大多数现有的方法依赖于多层感知器（mlp），但这些模型在不同的部署条件下往往存在精度有限和泛化能力差的问题。Kolmogorov-Arnold网络（KANs），其基于b样条的基函数，更好地捕捉复杂的非线性关系，同时降低过拟合风险。然而，原始的KANs仍然会产生很高的计算成本。为了解决这个问题，我们提出了WiKAN(WiFi KAN)，这是一种轻量级的基于KAN的室内WiFi定位模型。WiKAN通过将网络结构简化为两个KANLinear层，并用重构基函数优化矩阵乘法取代参数密集型操作，从而降低了计算复杂度。与传统的基系数计算相比，矩阵运算可以在现代硬件上更快地进行推理并提高可扩展性。此外，WiKAN通过可学习的线性组合集成了SiLU和b样条激活，平衡了光滑逼近和非线性表示。在UJIIndoorLoc、Tampere和JARIL三个基准数据集上的实验表明，WiKAN在MLP和标准KAN模型上都取得了卓越的性能：超过99.9%的建筑精度，高达100%的楼层分类，平均定位误差降低到5.91米。此外，运行时分析和参数计数比较证实了模型的计算效率。代码可在https://github.com/gyl555666/WiKAN公开获取。

{"title":"WiKAN: Lightweight Kolmogorov–Arnold Networks for accurate indoor WiFi localization","authors":"Yunlong Gu , Meng Xu , Jiguang Li , Qilei Li , Zhao Huang , Mengshan Li , Lixin Guan , Mikko Valkama","doi":"10.1016/j.pmcj.2025.102121","DOIUrl":"10.1016/j.pmcj.2025.102121","url":null,"abstract":"<div><div>With the growing demand for location-based services, WiFi localization plays a critical role in indoor environments. While most existing methods rely on Multi-Layer Perceptrons (MLPs), these models often suffer from limited accuracy and poor generalization across diverse deployment conditions. Kolmogorov–Arnold Networks (KANs), with their B-spline-based basis functions, better capture complex nonlinear relationships while reducing overfitting risks. However, original KANs still incur high computational costs. To address this, we propose WiKAN(WiFi KAN), a lightweight KAN-based model for indoor WiFi localization. WiKAN reduces computational complexity by simplifying the network structure to just two KANLinear layers and replacing parameter-intensive operations with optimized matrix multiplications using reconstructed basis functions. Compared to conventional computation of basis coefficients, matrix operations enable faster inference on modern hardware and improve scalability. Furthermore, WiKAN integrates SiLU and B-spline activations through a learnable linear combination, balancing smooth approximation and nonlinear representation. Experiments on three benchmark datasets (UJIIndoorLoc, Tampere, and JARIL) demonstrate that WiKAN achieves superior performance to both MLP and standard KAN models: over 99.9% building accuracy, up to 100% floor classification, and average positioning error reduced to 5.91 meters. Additionally, runtime analysis and parameter count comparisons confirm the model’s computational efficiency. Code is publicly available at: <span><span>https://github.com/gyl555666/WiKAN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102121"},"PeriodicalIF":3.5,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268085","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Construction of Wi-Fi fingerprint database based on WGAN-PSO: A method to alleviate signal sparsity and environmental noise 基于WGAN-PSO的Wi-Fi指纹库构建：一种缓解信号稀疏性和环境噪声的方法

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-10-01 DOI: 10.1016/j.pmcj.2025.102120

Heng Xu, Fanyu Meng, Long Sun, Hui Shao, Cheng Wang

To address signal sparsity and environmental noise in offline fingerprint database construction, this paper proposes a collaborative optimization method that integrates Wasserstein Generative Adversarial Network (WGAN) and Particle Swarm Optimization (PSO). First, the target area is gridded, and mobile-collected coordinates/RSSI data form the base fingerprint database. WGAN then expands the fingerprints of the sparse region under geographic boundary constraints, enhancing data diversity and spatial coverage. Finally, PSO optimizes the parameters of the path loss model through a comprehensive objective function that minimizes the RSSI estimation error. The experiments were carried out in two scenarios, in which 4 AP nodes were deployed, respectively. Data from 114 and 99 reference points were collected, generating 338,710 and 517,332 fingerprint data entries, respectively. The results demonstrate that the optimized database retains the original data features while reducing the positioning error caused by multipath effects and signal fading. Compared to traditional methods, RMSE is improved by 7.16% and 2.46% in two distinct scenarios, validating the efficacy of the proposed co-optimization framework.

针对离线指纹数据库构建中的信号稀疏性和环境噪声问题，提出了一种融合Wasserstein生成对抗网络（WGAN）和粒子群算法（PSO）的协同优化方法。首先，对目标区域进行网格化，并将手机采集的坐标/RSSI数据组成基本指纹数据库。然后，WGAN在地理边界约束下扩展稀疏区域的指纹，增强数据的多样性和空间覆盖。最后，粒子群算法通过综合目标函数对路径损失模型的参数进行优化，使RSSI估计误差最小。实验分为两种场景，分别部署4个AP节点。收集了114个和99个参考点的数据，分别生成了338,710和517,332个指纹数据条目。结果表明，优化后的数据库在保留原始数据特征的同时，减小了多径效应和信号衰落引起的定位误差。与传统方法相比，在两种不同的场景下，RMSE分别提高了7.16%和2.46%，验证了所提出的协同优化框架的有效性。

{"title":"Construction of Wi-Fi fingerprint database based on WGAN-PSO: A method to alleviate signal sparsity and environmental noise","authors":"Heng Xu, Fanyu Meng, Long Sun, Hui Shao, Cheng Wang","doi":"10.1016/j.pmcj.2025.102120","DOIUrl":"10.1016/j.pmcj.2025.102120","url":null,"abstract":"<div><div>To address signal sparsity and environmental noise in offline fingerprint database construction, this paper proposes a collaborative optimization method that integrates Wasserstein Generative Adversarial Network (WGAN) and Particle Swarm Optimization (PSO). First, the target area is gridded, and mobile-collected coordinates/RSSI data form the base fingerprint database. WGAN then expands the fingerprints of the sparse region under geographic boundary constraints, enhancing data diversity and spatial coverage. Finally, PSO optimizes the parameters of the path loss model through a comprehensive objective function that minimizes the RSSI estimation error. The experiments were carried out in two scenarios, in which 4 AP nodes were deployed, respectively. Data from 114 and 99 reference points were collected, generating 338,710 and 517,332 fingerprint data entries, respectively. The results demonstrate that the optimized database retains the original data features while reducing the positioning error caused by multipath effects and signal fading. Compared to traditional methods, RMSE is improved by 7.16% and 2.46% in two distinct scenarios, validating the efficacy of the proposed co-optimization framework.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102120"},"PeriodicalIF":3.5,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Home activity recognition using infrequently-monitored HEMS Data 使用不经常监测的医疗卫生系统数据进行家庭活动识别

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-09-27 DOI: 10.1016/j.pmcj.2025.102119

Fukuharu Tanaka , Teruhiro Mizumoto , Hirozumi Yamaguchi

This paper proposes a method for estimating household activities based only on the cumulative power consumption data obtained from the HEMS home distribution board every 30 min. The proposed method predicts the activity of each 30 min timeslot from the eight activity labels; household-level waking-up, household-level going-to-bed, room-level waking-up, room-level going-to-bed, cooking, laundry, dishwashing, and bathing. For the prediction, we first identify the branch circuit that is strongly correlated with each activity label and detect the turn-on/off of home appliances on the circuit to detect those activities. We also incorporate machine learning for estimating the other activities based on the circuit’s time series of power consumption. Furthermore, to cope with the difference among households, we apply transfer learning to the constructed model. In collaboration with a Japanese home builder, we conducted an experiment on five households using their HEMS data. In parallel, we obtained verifiable activity labels as our ground truth by the installation of specialized sensors in the respective homes. Under a ±30 min tolerance (i.e. allowing a prediction in the immediately preceding or following half-hour slot), our model achieved an average F1 score of 0.689 across all activities. We also confirmed that transfer learning improved the F1 score of each activity recognition and achieved an average improvement of 0.260 in household-level waking-up, household-level going-to-bed, room-level waking-up, room-level going-to-bed, and bathing activities.

本文提出了一种仅基于HEMS家庭配电板每30分钟获得的累计电力消耗数据来估计家庭活动的方法。该方法从8个活动标签中预测每个30分钟时间段的活动；家庭级别的起床，家庭级别的睡觉，房间级别的起床，房间级别的睡觉，做饭，洗衣，洗碗，洗澡。为了预测，我们首先确定与每个活动标签密切相关的分支电路，并检测电路上家用电器的开关以检测这些活动。我们还结合了机器学习来估计基于电路功耗时间序列的其他活动。此外，为了处理家庭之间的差异，我们将迁移学习应用于所构建的模型。我们与一家日本房屋建筑商合作，对五户家庭进行了实验，使用他们的医疗卫生系统数据。与此同时，我们通过在各自的家庭中安装专门的传感器，获得了可验证的活动标签作为我们的基础事实。在±30分钟的误差范围内（即允许在之前或之后的半小时时段进行预测），我们的模型在所有活动中获得了0.689的平均F1分数。我们还证实，迁移学习提高了各项活动识别的F1得分，在家庭级起床、家庭级上床睡觉、房间级起床、房间级上床睡觉和洗澡活动中平均提高了0.260。

{"title":"Home activity recognition using infrequently-monitored HEMS Data","authors":"Fukuharu Tanaka , Teruhiro Mizumoto , Hirozumi Yamaguchi","doi":"10.1016/j.pmcj.2025.102119","DOIUrl":"10.1016/j.pmcj.2025.102119","url":null,"abstract":"<div><div>This paper proposes a method for estimating household activities based only on the cumulative power consumption data obtained from the HEMS home distribution board every 30 min. The proposed method predicts the activity of each 30 min timeslot from the eight activity labels; household-level waking-up, household-level going-to-bed, room-level waking-up, room-level going-to-bed, cooking, laundry, dishwashing, and bathing. For the prediction, we first identify the branch circuit that is strongly correlated with each activity label and detect the turn-on/off of home appliances on the circuit to detect those activities. We also incorporate machine learning for estimating the other activities based on the circuit’s time series of power consumption. Furthermore, to cope with the difference among households, we apply transfer learning to the constructed model. In collaboration with a Japanese home builder, we conducted an experiment on five households using their HEMS data. In parallel, we obtained verifiable activity labels as our ground truth by the installation of specialized sensors in the respective homes. Under a ±30 min tolerance (i.e. allowing a prediction in the immediately preceding or following half-hour slot), our model achieved an average F1 score of 0.689 across all activities. We also confirmed that transfer learning improved the F1 score of each activity recognition and achieved an average improvement of 0.260 in household-level waking-up, household-level going-to-bed, room-level waking-up, room-level going-to-bed, and bathing activities.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102119"},"PeriodicalIF":3.5,"publicationDate":"2025-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145227084","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

On-device indoor place prediction using WiFi-RTT and inertial sensors 使用WiFi-RTT和惯性传感器的设备室内位置预测

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-09-22 DOI: 10.1016/j.pmcj.2025.102118

Pritam Sen , Xiaopeng Jiang , Qiong Wu , Manoop Talasila , Wen-Ling Hsu , Cristian Borcea

High-accuracy and low-latency indoor place prediction for mobile users can enable a wide range of applications for domains such as assisted living and smart homes. In this paper, we propose GoPlaces, a practical indoor place prediction system that works on mobile devices without requiring any new infrastructure. GoPlaces does not rely on servers or specialized localization infrastructure, except for a single cheap off-the-shelf WiFi access point that supports ranging with Round Trip Time (RTT) protocol. GoPlaces enables personalized place naming and prediction, and it protects users’ location privacy. It fuses inertial sensor data with distances estimated using the WiFi-RTT protocol to predict the indoor places a user will visit. GoPlaces employs an attention-based BiLSTM model to detect user’s current trajectory, which is then used together with historical information stored in a prediction tree to infer user’s future places. We implemented GoPlaces in Android and evaluated it in several indoor spaces. The experimental results demonstrate prediction accuracy as high as 86%. Furthermore, they show GoPlaces is feasible in real life because it has low latency and low resource consumption on the phones.

移动用户的高精度和低延迟室内位置预测可以为辅助生活和智能家居等领域提供广泛的应用。在本文中，我们提出了GoPlaces，这是一个实用的室内位置预测系统，可以在移动设备上工作，而不需要任何新的基础设施。GoPlaces不依赖于服务器或专门的定位基础设施，除了一个廉价的现成WiFi接入点，它支持往返时间（RTT）协议。GoPlaces支持个性化的地点命名和预测，并保护用户的位置隐私。它将惯性传感器数据与使用WiFi-RTT协议估计的距离融合在一起，以预测用户将访问的室内位置。GoPlaces采用基于注意力的BiLSTM模型来检测用户当前的轨迹，然后将其与存储在预测树中的历史信息一起使用，以推断用户未来的位置。我们在Android上实现了GoPlaces，并在几个室内空间中进行了评估。实验结果表明，预测准确率高达86%。此外，他们还表明，GoPlaces在现实生活中是可行的，因为它在手机上具有低延迟和低资源消耗。

{"title":"On-device indoor place prediction using WiFi-RTT and inertial sensors","authors":"Pritam Sen , Xiaopeng Jiang , Qiong Wu , Manoop Talasila , Wen-Ling Hsu , Cristian Borcea","doi":"10.1016/j.pmcj.2025.102118","DOIUrl":"10.1016/j.pmcj.2025.102118","url":null,"abstract":"<div><div>High-accuracy and low-latency indoor place prediction for mobile users can enable a wide range of applications for domains such as assisted living and smart homes. In this paper, we propose GoPlaces, a practical indoor place prediction system that works on mobile devices without requiring any new infrastructure. GoPlaces does not rely on servers or specialized localization infrastructure, except for a single cheap off-the-shelf WiFi access point that supports ranging with Round Trip Time (RTT) protocol. GoPlaces enables personalized place naming and prediction, and it protects users’ location privacy. It fuses inertial sensor data with distances estimated using the WiFi-RTT protocol to predict the indoor places a user will visit. GoPlaces employs an attention-based BiLSTM model to detect user’s current trajectory, which is then used together with historical information stored in a prediction tree to infer user’s future places. We implemented GoPlaces in Android and evaluated it in several indoor spaces. The experimental results demonstrate prediction accuracy as high as 86%. Furthermore, they show GoPlaces is feasible in real life because it has low latency and low resource consumption on the phones.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102118"},"PeriodicalIF":3.5,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145158997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Coordinated Q-learning based Multi-hop Routing for UAV-assisted communication 基于协同q学习的无人机辅助通信多跳路由

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-09-15 DOI: 10.1016/j.pmcj.2025.102105

N.P. Sharvari , Dibakar Das , Jyotsna Bapat , Debabrata Das

Unmanned Aerial Vehicle (UAV) assisted communication is gaining prominence as a vital solution for establishing effective emergency communication during disaster management operations. UAVs are essential for enhancing and expanding communication systems, acting as relays to boost data transmission to ground stations, extend network coverage, and provide connectivity. However, the dynamic and resource-limited nature of aerial networks necessitates robust routing mechanisms to facilitate seamless data dissemination. While existing Q-learning-based routing protocols are adaptive to changing network conditions and resilient to failures, they often lead to suboptimal network-wide decisions due to UAVs operating independently, each maximizing its gains. This paper proposes a novel Coordinated Q-learning-based Multi-hop Routing (CQMR) algorithm for multi-UAV networks. To the best of our knowledge, this is the first time a routing algorithm introduces UAV coordination for data routing through utility function approximation with a message-passing scheme, enabling the selection of globally optimal joint actions. This novel approach meticulously considers a comprehensive set of parameters for data routing, including minimizing the expected number of hops to the destination, monitoring energy usage, maintaining network connectivity, preventing UAV collisions, and supporting adaptive network reorganization. This integrated consideration of multiple factors positions the proposed solution as superior to existing work, offering a uniquely robust and highly effective strategy for UAV-assisted communication in dynamic, resource-constrained environments, such as emergency scenarios. CQMR builds upon and extends the Improved Q-learning-based Multi-hop Routing (IQMR) algorithm, demonstrating a 12.47% increase in energy efficiency and a 13.34% higher success rate in data transmission compared to IQMR while requiring 40% fewer hops to reach the destination.

无人机（UAV）辅助通信作为在灾害管理行动中建立有效应急通信的重要解决方案，正日益受到重视。无人机对于增强和扩展通信系统至关重要，充当中继器，促进数据传输到地面站，扩展网络覆盖范围，并提供连接。然而，空中网络的动态性和资源有限性需要强大的路由机制来促进无缝数据传播。虽然现有的基于q学习的路由协议可以适应不断变化的网络条件，并且能够适应故障，但由于无人机独立运行，每个无人机的收益最大化，它们通常会导致网络范围内的次优决策。针对多无人机网络，提出了一种基于协同q学习的多跳路由算法。据我们所知，这是路由算法首次通过消息传递方案的效用函数近似引入无人机协调数据路由，从而实现全局最优联合动作的选择。这种新颖的方法仔细考虑了数据路由的一组综合参数，包括最小化到目的地的预期跳数，监控能源使用，维护网络连接，防止无人机碰撞，并支持自适应网络重组。综合考虑多种因素，提出的解决方案优于现有工作，为在动态、资源受限环境（如紧急情况）下的无人机辅助通信提供了一种独特、强大和高效的策略。CQMR建立并扩展了基于改进q学习的多跳路由（IQMR）算法，与IQMR相比，能效提高了12.47%，数据传输成功率提高了13.34%，到达目的地所需的跳数减少了40%。

{"title":"Coordinated Q-learning based Multi-hop Routing for UAV-assisted communication","authors":"N.P. Sharvari , Dibakar Das , Jyotsna Bapat , Debabrata Das","doi":"10.1016/j.pmcj.2025.102105","DOIUrl":"10.1016/j.pmcj.2025.102105","url":null,"abstract":"<div><div>Unmanned Aerial Vehicle (UAV) assisted communication is gaining prominence as a vital solution for establishing effective emergency communication during disaster management operations. UAVs are essential for enhancing and expanding communication systems, acting as relays to boost data transmission to ground stations, extend network coverage, and provide connectivity. However, the dynamic and resource-limited nature of aerial networks necessitates robust routing mechanisms to facilitate seamless data dissemination. While existing Q-learning-based routing protocols are adaptive to changing network conditions and resilient to failures, they often lead to suboptimal network-wide decisions due to UAVs operating independently, each maximizing its gains. This paper proposes a novel Coordinated Q-learning-based Multi-hop Routing (CQMR) algorithm for multi-UAV networks. To the best of our knowledge, this is the first time a routing algorithm introduces UAV coordination for data routing through utility function approximation with a message-passing scheme, enabling the selection of globally optimal joint actions. This novel approach meticulously considers a comprehensive set of parameters for data routing, including minimizing the expected number of hops to the destination, monitoring energy usage, maintaining network connectivity, preventing UAV collisions, and supporting adaptive network reorganization. This integrated consideration of multiple factors positions the proposed solution as superior to existing work, offering a uniquely robust and highly effective strategy for UAV-assisted communication in dynamic, resource-constrained environments, such as emergency scenarios. CQMR builds upon and extends the Improved Q-learning-based Multi-hop Routing (IQMR) algorithm, demonstrating a 12.47% increase in energy efficiency and a 13.34% higher success rate in data transmission compared to IQMR while requiring 40% fewer hops to reach the destination.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102105"},"PeriodicalIF":3.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107510","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

IDENTIFY: Intelligent device identification using device fingerprints and machine learning 识别：使用设备指纹和机器学习的智能设备识别

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-09-15 DOI: 10.1016/j.pmcj.2025.102103

Liwei Liu , Muhammad Ajmal Azad , Harjinder Lallie , Hany Atlam

The Internet of Things (IoT) consists of a rapidly growing network of heterogeneous devices that autonomously monitor, collect, and exchange data across a wide range of application domains. The rapid increase of IoT devices highlighted the importance of scalable, secure, and adaptive network management strategies for dynamic networks. A key challenge in this context is the automatic identification of devices, which is critical for detecting and mitigating malicious devices that can compromise network integrity. Accurate device identification strengthens the security of dynamic IoT environments by facilitating early detection of anomalous or adversarial traffic. Device fingerprinting offers a non-intrusive solution by leveraging protocol and traffic characteristics, without relying on vendor-specific identifiers. In this work, we propose a lightweight and efficient framework for IoT device identification based on machine learning. Our model utilises a Random Forest classifier in conjunction with a data-driven feature selection strategy that emphasises low-overhead features derived from packet headers and traffic flow statistics. The proposed approach achieves high classification performance, attaining 97.32% accuracy in identifying general device categories and 94.39% accuracy for specific device types. It also demonstrates approximately a 40% improvement in computational efficiency compared to traditional classifiers, making it well-suited for deployment in resource-constrained edge environments. We evaluate the model under various real-world conditions, including spatiotemporal traffic variations, changes in operational modes, and different sampling intervals. Comparative experiments with established classifiers—such as J48, SMO, BayesNet, and Naive Bayes—are performed using standard metrics, including precision, recall, F1-score, and inference latency. Our approach strengthens network security by automatically identifying and classifying IoT devices in dynamic, heterogeneous environments. It is lightweight, scalable, and well-suited for deployment in resource-constrained IoT scenarios.

物联网（IoT）由快速增长的异构设备网络组成，这些设备在广泛的应用领域中自主监控、收集和交换数据。物联网设备的快速增长凸显了可扩展、安全和自适应网络管理策略对动态网络的重要性。在这种情况下的一个关键挑战是设备的自动识别，这对于检测和减轻可能危及网络完整性的恶意设备至关重要。准确的设备识别通过促进早期发现异常或敌对流量来增强动态物联网环境的安全性。设备指纹识别通过利用协议和流量特征提供了一种非侵入性的解决方案，而不依赖于特定于供应商的标识符。在这项工作中，我们提出了一个基于机器学习的轻量级高效物联网设备识别框架。我们的模型将随机森林分类器与数据驱动的特征选择策略相结合，该策略强调来自数据包头和交通流量统计的低开销特征。该方法具有较高的分类性能，识别一般设备类别的准确率为97.32%，识别特定设备类型的准确率为94.39%。与传统分类器相比，它的计算效率提高了大约40%，使其非常适合在资源受限的边缘环境中部署。我们在不同的现实条件下评估了该模型，包括时空交通变化、操作模式变化和不同的采样间隔。与已建立的分类器（如J48、SMO、BayesNet和朴素贝叶斯）进行比较实验，使用标准指标，包括精度、召回率、f1分数和推理延迟。我们的方法通过自动识别和分类动态异构环境中的物联网设备来增强网络安全性。它轻量级、可扩展，非常适合在资源受限的物联网场景中部署。

{"title":"IDENTIFY: Intelligent device identification using device fingerprints and machine learning","authors":"Liwei Liu , Muhammad Ajmal Azad , Harjinder Lallie , Hany Atlam","doi":"10.1016/j.pmcj.2025.102103","DOIUrl":"10.1016/j.pmcj.2025.102103","url":null,"abstract":"<div><div>The Internet of Things (IoT) consists of a rapidly growing network of heterogeneous devices that autonomously monitor, collect, and exchange data across a wide range of application domains. The rapid increase of IoT devices highlighted the importance of scalable, secure, and adaptive network management strategies for dynamic networks. A key challenge in this context is the automatic identification of devices, which is critical for detecting and mitigating malicious devices that can compromise network integrity. Accurate device identification strengthens the security of dynamic IoT environments by facilitating early detection of anomalous or adversarial traffic. Device fingerprinting offers a non-intrusive solution by leveraging protocol and traffic characteristics, without relying on vendor-specific identifiers. In this work, we propose a lightweight and efficient framework for IoT device identification based on machine learning. Our model utilises a Random Forest classifier in conjunction with a data-driven feature selection strategy that emphasises low-overhead features derived from packet headers and traffic flow statistics. The proposed approach achieves high classification performance, attaining 97.32% accuracy in identifying general device categories and 94.39% accuracy for specific device types. It also demonstrates approximately a 40% improvement in computational efficiency compared to traditional classifiers, making it well-suited for deployment in resource-constrained edge environments. We evaluate the model under various real-world conditions, including spatiotemporal traffic variations, changes in operational modes, and different sampling intervals. Comparative experiments with established classifiers—such as J48, SMO, BayesNet, and Naive Bayes—are performed using standard metrics, including precision, recall, F1-score, and inference latency. Our approach strengthens network security by automatically identifying and classifying IoT devices in dynamic, heterogeneous environments. It is lightweight, scalable, and well-suited for deployment in resource-constrained IoT scenarios.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"114 ","pages":"Article 102103"},"PeriodicalIF":3.5,"publicationDate":"2025-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107447","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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