Search Results (1,083)

Unmanned aerial vehicles (UAVs) have become essential cyber-physical platforms for applications such as surveillance, infrastructure inspection, emergency response, and intelligent transportation. However, their tight coupling among sensing, communication, control, actuation, and swarm coordination also exposes them to sophisticated cyber-physical attacks that are difficult to detect using conventional intrusion detection systems. Existing machine learning, deep learning, graph-based, and federated intrusion detection approaches generally rely on statistical feature representations or temporal patterns, providing limited capability to model causal dependencies among interacting UAV subsystems and to generalize across heterogeneous operating environments. To address these limitations, this paper proposes Fed-CGIDS-UAV, a federated causal graph learning framework for cross-domain intrusion detection in cyber-physical UAV networks. The proposed framework models each telemetry window as a typed causal graph in which nodes represent navigation, sensing, communication, control, actuation, and swarm states, while directed edges capture stable operational dependencies. Intrusions are detected by identifying violations of these learned causal relationships, and the framework provides interpretable node-edge explanations to support root-cause analysis. Furthermore, federated learning enables collaborative model training across distributed UAV clients without sharing raw telemetry, thereby preserving data privacy while improving robustness under heterogeneous operating conditions. The proposed framework was implemented and experimentally evaluated in a controlled simulation environment covering four UAV operating domains and six representative attack classes. All experiments were repeated over five independent runs using different random seeds, and the reported results correspond to the measured average performance. The proposed framework was implemented using Python 3.12 (Python Software Foundation, Wilmington, DE, USA) and PyTorch 2.3 (Meta Platforms, Menlo Park, CA, USA). UAV flight data were generated using Microsoft AirSim 1.9.1 (Microsoft Corporation, Redmond, WA, USA), integrated with PX4 Autopilot v1.14 (Dronecode Foundation, San Francisco, CA, USA) and Gazebo Sim 11 (Open Source Robotics Foundation, Mountain View, CA, USA). Within this simulation-based evaluation, Fed-CGIDS-UAV achieved an accuracy of 0.968, an F1-score of 0.956, and an internal–external stability gap (IESG) of 0.028, outperforming conventional machine learning, deep learning, graph-based, and centralized causal baselines while maintaining competitive computational latency. Although these results demonstrate the effectiveness of the proposed framework under controlled simulation conditions, validation using real-flight UAV telemetry remains an important direction for future research. These results demonstrate that integrating causal graph learning with federated optimization provides an effective and interpretable solution for privacy-preserving intrusion detection in heterogeneous cyber-physical UAV environments. Full article

Accurate and robust localization remains the fundamental bottleneck for truly autonomous robotic systems, despite decades of progress in probabilistic estimation and SLAM. This paper provides a critical and comprehensive review of mobile robot localization across sensing modalities, estimation paradigms, and deployment domains, covering ground, aerial, and underwater platforms. Beyond a descriptive survey, we explicitly analyze the limitations and trade-offs of existing approaches. We introduce an updated taxonomy that spans classical proprioceptive and exteroceptive sensors, emerging technologies such as 4D imaging radar and event cameras, and infrastructure-based positioning systems including GNSS and Ultra-Wideband. We revisit the evolution of localization algorithms, from Bayesian filtering techniques (EKF, UKF, and particle filters) to modern graph-based SLAM frameworks and tightly coupled multi-sensor fusion systems. Particular emphasis is placed on the recent paradigm shift toward learning-based and neural implicit approaches, including NeRF-SLAM and Gaussian Splatting, highlighting both their transformative potential and their current impracticality for real-time deployment. Unlike previous surveys, this work provides a unified cross-domain perspective while critically examining scalability, robustness, computational cost, and real-world deployability. We identify key unresolved challenges, including long-term consistency, operation in degraded environments, and the integration of semantic understanding into localization pipelines. Furthermore, we propose standardizing evaluation metrics with a formal Trajectory Completeness formulation to expose tracking brittleness. Finally, we outline future research directions toward resilient, certifiable, and truly autonomous localization systems, emphasizing the critical transition from passive estimation to Active SLAM in unstructured environments. Full article

Background/Objectives: Speech disorders caused by neurological, articulatory, cognitive, and behavioral impairments require accurate multimodal analysis frameworks capable of jointly understanding acoustic abnormalities and neurolinguistic inconsistencies for reliable computer-assisted speech disorder assessment. Recent multimodal deep learning approaches have utilized speech signals, linguistic transcripts, attention mechanisms, and contextual fusion strategies to improve disordered speech analysis and intelligent disorder prediction. However, existing methods frequently suffer from insufficient temporal synchronization, ineffective local–global feature learning, multimodal redundancy, poor interpretability, and reduced robustness under heterogeneous speech disorders. To address these challenges, Methods: The proposes NSX-Net (NeuroSpeech Explainable Network), an intelligent multimodal deep learning framework for speech disorder classification using Acoustic Speech Signal Data and Neurolinguistic Text/Linguistic Data. Initially, the Adaptive Speech Refinement Module (ASRM) performs noise removal, silence elimination, signal normalization, spectrogram generation, MFCC extraction, and transcript preprocessing to improve multimodal speech consistency. Subsequently, the Hierarchical Multimodal Feature Learning Unit (HMFLU) extracts discriminative feature representations through the Cross-Domain Representation Encoder (CDRE), Fine-Grained Speech Pattern Analyzer (FGSPA), and Global Sequential Dependency Learner (GSDL) for capturing both local articulation abnormalities and long-range semantic dependencies. Furthermore, the Dual-Path Attention Enhancement Block (DPAEB) emphasizes clinically important disordered speech regions using adaptive local–global attention mechanisms, while the Temporal Resolution Synchronization Module (TRSM) aligns rhythm-level and beat-level multimodal contextual structures to improve temporal consistency and suppress noise disturbances. Results: Experimental evaluation across six publicly available multimodal speech disorder datasets demonstrated that NSX-Net achieved superior performance, with 99.27% Accuracy, 99.11% Precision, 98.97% Recall, 99.02% F1-Score, and 99.41% AUC, significantly outperforming existing state-of-the-art frameworks. Conclusions: Finally, optimized multimodal contextual representations are forwarded into a Softmax classification layer for intelligent multiclass speech disorder prediction with improved interpretability and potential clinical applicability. Full article

Externally oriented corporate social responsibility (ESR), defined here as the community- and environment-facing domain of CSR, may shape how residents evaluate firms in environmentally sensitive industries. This study examines the predictive association between community-perceived ESR and corporate image using a stratified random sample of 384 adults from the Chimbote metropolitan area, Peru. A 21-item ESR instrument and a 20-item corporate-image instrument were analyzed using partial least squares structural equation modeling (PLS-SEM), with corporate image specified as a Type I reflective-reflective higher-order construct comprising reputation, trust, perceived responsibility, social commitment, and familiarity/recognition. ESR was positively associated with corporate image (β = 0.519, t = 13.934, p < 0.001, 95% CI [0.449, 0.595]), explaining 26.9% of its variance (R² = 0.269; f² = 0.368; Q² = 0.265). The five first-order image dimensions displayed substantial empirical overlap, supporting a higher-order representation in this sample. Common-method-bias diagnostics, confirmatory factor analysis, covariance-based SEM, and robust OLS estimates supported the stability of the association. The findings suggest that consistent, visible, and community-oriented responsibility practices are associated with more favorable corporate evaluations in a high-salience fishing context. Because the design is cross-sectional, the results should be interpreted as predictive associations rather than causal effects and should not be generalized beyond comparable settings without replication. Full article

Learning analytics increasingly relies on large-scale educational activity logs to understand student behavior and support intelligent decision making. However, many existing approaches primarily exploit either semantic information or sequential dependencies independently while overlooking the joint influence of temporal context, behavioral transitions, and probabilistic reliability. This paper addresses the problem of next-event prediction in student activity streams by proposing the Time-Aware Student Event Model (TSEM), a unified framework that integrates semantic representation learning, temporal contextual encoding, and sequential transition modeling within a hybrid neural–probabilistic architecture. The proposed framework is evaluated using a large-scale institutional dataset containing approximately 2.49 million student activity events collected over four academic years, from which nearly 900,000 temporally ordered next-event prediction pairs are constructed. Extensive experiments compare the proposed approach against strong statistical and deep learning baselines, including Markov models, TF-IDF + Logistic Regression, LSTM, GRU, and Transformer architectures, under both random and temporal evaluation protocols. Furthermore, cross-domain validation is performed on the public EdNet benchmark to assess generalization across heterogeneous educational environments. Experimental results demonstrate that jointly modeling semantic, temporal, and sequential information consistently improves predictive performance, robustness to temporal distribution shifts, and probability calibration while maintaining competitive generalization across independent datasets. Ablation studies, statistical significance analysis, and calibration evaluation further confirm the contribution of each modeling component and the reliability of the proposed framework. These findings demonstrate that TSEM provides a robust and practically deployable solution for intelligent educational analytics, supporting applications such as adaptive learning recommendation, early student intervention, learning analytics, and confidence-aware educational decision support. Full article

Multimodal remote sensing data provide important data support for large-scale, all-weather monitoring of offshore oil and gas platforms. However, differences in imaging modalities and acquisition conditions introduce significant domain shift, which limits the generalization capability of object detection models in unseen target domains. Furthermore, most existing multimodal detection methods typically rely on paired and strictly registered data, thereby restricting their applicability in practical scenarios. To overcome these challenges, an end-to-end method, termed OG-MLDD, is proposed for cross-modal offshore oil and gas platform detection. Without requiring multimodal image registration, the proposed method integrates meta-learning with domain discriminator (DD) to encourage the extraction of semantically discriminative representations that remain invariant across different domains. Specifically, a dual-gradient-descent-based meta-learning strategy is first introduced to jointly optimize the meta-train and meta-test objectives, encouraging the model to learn generalized and discriminative feature representations across different imaging modalities. Subsequently, a dynamic-weighted gradient reversal layer (DWGRL) is embedded into DD to guide adversarial feature learning between different domains, thereby promoting the acquisition of domain-robust representations and supplying global supervisory cues throughout the training process. Finally, a multi-scale feature aggregation (MFA) module is proposed to effectively integrate fine-grained local details with multiple receptive fields and high-level contextual information, thereby refining the fused feature representations and enhancing the representation capability of multi-scale offshore platform objects. Experimental results on a real-world dataset demonstrate that, even without using target domain data during training, the proposed method achieves superior detection performance and robust cross-domain generalization. Compared with existing methods, the full method improves detection accuracy by approximately 8 percentage points with a standard deviation of only 0.6, highlighting its superior detection performance and generalization ability. Full article

Credit risk assessment is a core component of financial decision-making. This study develops an explainable machine learning framework for modeling loan approval decisions on heterogeneous tabular data, centered on a Cross-Attentional Tabular Transformer that applies bidirectional cross-attention between numerical and categorical feature groups. The prediction target is historical loan-approval status, treated as a proxy for, not a direct measure of, borrower default risk; a supplementary validation on a dataset with an authentic default label is also reported. Class imbalance is addressed through focal loss, and post hoc interpretability is provided through SHAP analysis. Three classifiers, Random Forest, Gradient Boosting, and the proposed transformer, are evaluated on a 5000-sample credit dataset using accuracy, precision, recall, F1-score, ROC-AUC, and average precision. Gradient Boosting achieves the best performance (accuracy 0.9640, F1-score 0.9189), with Random Forest comparable; the proposed transformer reaches 0.9530 accuracy and 0.8949 F1, without surpassing the ensembles and at substantially higher computational cost. A five-split robustness comparison additionally evaluates XGBoost, LightGBM, CatBoost, and calibrated logistic regression: all three Gradient-Boosting variants and both classical ensembles exceed the transformer’s performance on every metric, while calibrated logistic regression does not. The evaluated baseline set excludes deep tabular architectures such as TabNet, FT-Transformer, SAINT, and TabPFN-style methods. Across the three primary classifiers, SHAP identifies credit score, employment status, and income as the dominant features, consistent with domain expectations. The results characterize the observed performance–efficiency trade-off between ensemble methods and attention-based tabular learning under the evaluated data conditions. Full article

Financial statement fraud inflicts large and recurring losses on capital markets, yet the dominant detection paradigm still relies on single, black-box classifiers (e.g., RUSBoost) trained on structured accounting ratios alone. Two limitations follow: (i) the rich, unstructured Management Discussion and Analysis (MD&A) narrative of the 10-K filing is discarded, and (ii) the resulting scores are difficult for auditors to trust because they carry no transparent, standards-aligned rationale. Recent large language model (LLM) systems have shown that multi-agent collaboration is more robust than a single LLM for anomaly detection, but no study has systematically transferred this paradigm to listed-company statement fraud. We propose FraudDebate-Agent, a four-role multi-agent system in which a Quantitative Analyst agent scores 28 raw accounting items and 14 ratios with gradient-boosted and tabular attention models, a Narrative Auditor agent quantifies tone, linguistic uncertainty, and year-over-year textual novelty of the MD&A with FinBERT, and an Industry Peer agent uses retrieval-augmented generation to measure industry-relative anomaly. A Critic–Debate agent then orchestrates a pair-wise Evidence-based Multi-Agent Debate (EMAD) that reconciles disagreement across modalities and arbitrates a reconciled fraud-risk assessment, which is aggregated over a tri-modal evidence graph. Our contributions are as follows: (1) the first use of an evidence-grounded debate mechanism for accounting fraud, which materially reduces LLM hallucination; (2) a numerical–textual–peer evidence graph that fuses heterogeneous signals; and (3) an explainable report aligned with the PCAOB AS 2401 fraud-risk taxonomy. On AAER-labelled firm-years linked across a SEC financial dataset and EDGAR-CORPUS, FraudDebate-Agent improves the area under the ROC curve and the rare-event ranking metric NDCG@k over the strongest single-modality and single-LLM baselines while producing substantially more faithful explanations. We frame the system as a fraud-risk screening and risk-ranking tool for AAER-labelled misstatement risk rather than a determination of fraudulent intent. We report results over multiple seeds to reflect real-world stochasticity and discuss limitations and cross-domain applications. Full article

Background/Objectives: Symmetrical dual-site acetylcholinesterase (AChE) inhibitors offer a compelling strategy to mitigate mutational drug resistance, yet static modeling fails to capture induced-fit dynamics under mutational stress. Methods: Here, a 100,000-compound virtual library was filtered using a machine learning-based QSAR classification pipeline. A strict, empirically calibrated Jaccard applicability domain filter (AD = 0.823) eliminated topological anomalies, yielding a robust cross-validation accuracy (ROC-AUC: 0.80 ± 0.05; independent test MCC: 0.61). Multi-parameter ADMET and shape screening prioritized unique chemotypes to probe the 20 Å enzyme gorge. All-atom explicit-solvent molecular dynamics simulations were coupled with 150 ns enhanced-sampling Metadynamics along two orthogonal collective variables (gorge depth and ligand orientation) to map out the free energy surfaces under mutational stress. Results: Symmetrical probes suffered catastrophic unbinding upon anchor loss. Conversely, the symmetrical core of Lead Compound 1631 demonstrated extraordinary structural resilience. In silico site-directed mutagenesis (W86A and W286A) triggered a thermodynamic locking effect; the W86A mutant forced the complex into a deeper energetic well (ΔG_min = 9.23 ± 1.98 kJ/mol) than the wild-type state (5.26 ± 1.69 kJ/mol). MM/GBSA decomposition confirmed an active electrostatic-solvation compensation mechanism along a “solvation see-saw” diagonal (ΔΔG_total = +1.59 kcal/mol). Finally, Dynamic Cross-Correlation Matrix analysis quantified a mechanical inversion of the CAS-PAS axis into an anti-correlated clamping mode (−0.04) that locked the ligand bridge in place. Conclusions: These results demonstrate that symmetrical dual-site targeting, combined with dynamic thermodynamic locking, provides a resilient framework to overcome mutational resistance in AChE inhibitors. Full article

This paper presents a numerical-computational analysis of ${\mathsf{\ell }}_2$-regularized logistic learning for binary intrusion detection under heterogeneous datasets, class imbalance, and cross-dataset shift. Rather than proposing a new intrusion detection architecture, the study examines how numerical conditioning, feature scaling, feature set design, threshold selection, false negative behavior, false alarm behavior, and distribution shift affect operational detection performance. Experiments were conducted on CICIDS2017, UNSW-NB15, and CIRA-CIC-DoHBrw-2020 using reproducible train–validation–test protocols over five fixed random seeds. The numerical audit showed that standard scaling reduced the spectral condition number of traffic feature matrices by several orders of magnitude across datasets and feature configurations. However, scaling did not produce uniformly monotonic predictive gains: in some cases, raw feature optimization achieved comparable or higher F1-score, whereas scaled preprocessing produced more controlled false alarm behavior. In-domain experiments showed that dataset-specific features may improve ranking metrics such as area under the receiver-operating-characteristic curve (AUROC) or area under the precision–recall curve (AUPR) without necessarily improving thresholded operational metrics. Cross-dataset transfer experiments revealed strong source–target asymmetry, with transferred thresholds producing either near-zero positive detection or excessive false alarms. Additional robustness experiments with Random Forest and XGBoost improved in-domain F1-score and false negative rate (FNR), but did not eliminate off-domain degradation, with high FNR persisting under direct cross-dataset transfer. Finally, a Kolmogorov–Smirnov-based distribution shift analysis showed that in-domain discrepancies were small, whereas cross-dataset discrepancies were consistently large under common standardized traffic features. These findings suggest that numerical stability, ranking quality, thresholded detection performance, false negative and false alarm behavior, and distribution shift should be analyzed jointly when evaluating intrusion detection models. Full article

Background and Objectives: Evidence on the acute effects of caffeine in female team-sport athletes is limited. This study aimed to quantify the acute effects of caffeine on female team-sport athletes across physical performance, sport-specific performance, physiological responses, perceptual responses, and cognitive performance and determine moderators. Methods: PubMed and Web of Science were systematically searched for randomized, placebo-controlled crossover trials of acute, dose-defined caffeine in female team-sport athletes. Effect sizes were expressed as Hedges’ g and synthesized within each domain using a three-level CHE model with CR2 cluster-robust variance estimation and Satterthwaite degrees of freedom; 95% confidence intervals and prediction intervals were reported. Risk of bias (RoB 2), methodological quality (PEDro), and certainty of evidence (GRADE) were assessed. Results: Twenty-six studies were included, comprising 26 randomized, blind, placebo-controlled crossover trials. As a descriptive cross-domain summary, acute caffeine intake showed a small overall effect when all available domains were pooled (Hedges’ g = 0.24, 95% CI 0.13 to 0.35); however, because these domains represent different constructs, domain-specific estimates were considered the primary interpretable findings. For physical performance, caffeine showed a small favorable effect (g = 0.32, 95% CI 0.22 to 0.42), with statistical evidence for repeated sprint ability (g = 0.43), agility or change in direction (g = 0.42), sprint or speed performance (g = 0.39), anaerobic power (g = 0.30), and jumping performance (g = 0.29). For sport-specific performance, the pooled estimate indicated a small favorable effect but did not reach conventional statistical significance (g = 0.36, 95% CI: −0.01 to 0.73). Exploratory sub-indicator analyses, each based on only a few studies, suggested possible favorable directions for sport-specific locomotion or running performance (g = 0.46) and throwing or ball-speed outcomes (g = 0.19), whereas evidence was insufficient for shooting or scoring accuracy. Physiological outcomes were direction-aligned to the value conventionally regarded as favourable for each indicator (i.e., lower heart rate, lactate, and glucose), so that positive values denote the conventionally favourable direction; because the domain aggregates indicators of differing clinical desirability, its pooled value is a descriptive summary only and showed no clear domain-level direction (g = −0.05, 95% CI −0.31 to 0.21). The heart rate submetric showed a statistically detectable difference (g = −0.30), indicating that caffeine increased heart rate, the expected pharmacological response to a stimulant rather than an adverse effect; there were no clear effects on blood lactate or blood glucose. Perceptual responses improved modestly (g = 0.42), largely through reduced rating of perceived exertion (g = 0.35). Cognitive performance, based on only four studies with low degrees of freedom and wide uncertainty, showed a nonsignificant direction (g = 0.37) with no clear evidence for reaction time or cognitive accuracy, and these data should not be interpreted as showing that caffeine improves cognitive performance in female team-sport athletes. Exploratory moderator analyses did not provide statistically reliable evidence of moderation by caffeine dose, timing of intake, formulation or source, sport type, competitive level, habitual caffeine intake, or blinding status. Conclusions: Available evidence most consistently supports small, outcome-specific benefits of acute caffeine for physical performance and reduced perceived exertion. Evidence for sport-specific skills and cognitive outcomes remained uncertain, based on few studies per outcome and not supporting a general benefit. Current data do not support dose-, timing-, formulation-, habitual-intake-, sport-, or population-stratified recommendations. Future trials should be adequately powered, prospectively report menstrual-cycle phase, hormonal-contraceptive use, habitual caffeine intake, and adverse symptoms, and verify the integrity of blinding. Full article

Machine learning techniques for network-based intrusion detection systems (NIDS) have advanced considerably over the past decade. Still, improvements are inhibited by handcrafted feature pipelines, isolated public benchmark data, and evaluation procedures that do not reflect real-life deployment. AutoML, a branch of ML automating model selection, automated architecture search, and the creation of model pipelines, may help overcome these shortcomings. While numerous NIDS applications employing automated ML techniques have been proposed, and recent surveys have mapped the AutoML framework landscape for network intrusion detection, no existing review critically audits the evaluation practice of this literature: the quality of its benchmark datasets, the reproducibility of its reported results, and the realism of its deployment assumptions. This paper critically reviews 26 research works published between January 2023 and June 2026, collected via a two-phase structured search: a documented keyword search across five databases (Scopus, IEEE Xplore, Web of Science, ACM Digital Library, and Google Scholar), followed by full-text eligibility screening, citation chaining, and expert evaluation. Findings drawn from this collection capture trends observed among the selected studies, rather than reflecting the broader state of the field. Analysis of the corpus reveals that 88% of dataset-verified studies evaluate exclusively or partly on the legacy benchmark family (KDD-derived, CICIDS, UNSW-NB15, CIDDS), 21% evaluate on a single dataset only, and among attribute-verified studies only 32% release source code, 40% report statistical significance testing, and 36% include variance analysis, findings that collectively motivate the four contributions of this study. First, a recommended evaluation framework is proposed, addressing baseline parity, transparent search-space and budget reporting, nested cross-validation for selection-bias control, and stability reporting across multiple random seeds. Second, a dataset quality scoring framework is introduced, assessing five dimensions: overlap rate, duplication rate, label correctness, attack-type representativeness, and coverage of benign, IoT, and IIoT traffic. Third, a cross-domain justification is provided for neural architecture search (NAS) and meta-learning in NIDS, grounded in advances in federated NAS, out-of-distribution robustness, edge-constrained search cost reduction, and few-shot adaptation. Fourth, a structured research roadmap is outlined, targeting real-world validation, standardized benchmarks, curated datasets, resource-aware AutoML, and privacy-preserving federated NAS. In contrast to prior surveys of AutoML for network intrusion detection, which map frameworks and computational paradigms, this review contributes a formalized evaluation checklist, an explicit and partially empirically validated dataset quality scoring scheme, and evidence-based methodological guidance grounded in a transparent, fully enumerated study corpus. Full article

In Industrial Internet of Things (IIoT) scenarios, data privacy constraints and distribution discrepancies caused by time-varying working conditions hinder existing intelligent fault diagnosis models from maintaining robust generalization performance across different, particularly unknown, working conditions. To address this challenge, a federated generalization fault diagnosis method driven by statistical manifold generation is proposed. The method constructs a closed-loop collaborative strategy. Initially, individual users extract and upload representative statistical information (SI) as lightweight, privacy-preserving knowledge carriers. Subsequently, a global Gaussian mixture model coupled with a covariance expansion mechanism is established in the cloud to fit multi-source distributions and extrapolate uncertainty boundaries, thereby generating virtual SI. Finally, this virtual SI is assigned via a difference-aware mechanism and integrated locally using instance normalization to achieve domain-invariant augmented training. Extensive distributed collaborative fault diagnosis experiments conducted on rolling bearing and gearbox datasets demonstrate that, when facing completely unknown working conditions, the proposed method achieves an average diagnostic accuracy of over 85%, exceeding 90% in some tasks. Furthermore, the communication payload per round is merely 1.25 KB. While strictly preserving data privacy, the proposed method significantly enhances the cross-domain generalization capability of local models with minimal communication overhead, providing an efficient and robust collaborative intelligent diagnosis solution for resource-constrained IIoT edge devices. Full article

Background: The preservation of intrinsic capacity (IC) is a global public health priority, and digital health technologies offer promising avenues for managing functional decline. However, the longitudinal impact of specific smart monitoring devices on IC trajectories remains underexplored, and theoretical debates persist regarding whether digital interventions exacerbate health inequalities (the “digital divide”) or mitigate them (the “compensatory effect”). Methods: Using a balanced panel of 2432 older adults from the China Longitudinal Aging Social Survey (CLASS 2020–2023), we conducted wave-specific latent profile analyses based on five standardized WHO intrinsic-capacity domains. Modal class assignment and cross-wave profile matching were subsequently used to describe profile stability and transitions. The longitudinal protective association of baseline smart blood pressure (BP) monitor use with IC trajectories was evaluated using multinomial logistic regression with Inverse Probability Weighting (IPW). Exploratory structural path analysis was conducted to examine a potential mediating pathway. Results: The baseline latent profile analysis identified three profiles: Robust Capacity (80.2%), Cognitively Impaired (10.9%), and Sensory Impaired (8.8%). The Sensory Impaired class exhibited strong path dependence, with 62.79% of participants remaining in the same state over three years. Baseline estimates revealed that smart BP monitor usage was significantly associated with a reduced relative risk of transitioning into or remaining in the Sensory Impaired profile (RRR = 0.557, 95% CI: 0.325–0.955). The IPW analysis yielded a directionally consistent estimate (RRR = 0.696), although the association was attenuated and was not statistically significant (p = 0.162). Subgroup and pathway analyses yielded non-significant but suggestive findings, indicating that protective associations might be concentrated among vulnerable groups, particularly women and individuals with lower educational attainment. Conclusions: Smart physiological monitoring interventions demonstrate a longitudinal protective association against the persistence of physical frailty. Rather than definitively bridging the digital divide, targeted medical-grade digital tools may serve as compensatory resources for specific vulnerable subpopulations. These observational associations highlight the need for further quasi-experimental testing before integrating specific digital interventions into community-based chronic care ecosystems. Full article

Building materials and their spatial distribution play a significant role in determining the outcomes of human-caused and natural disasters in urban and peri-urban areas. However, building-level data on building and roofing materials are scarce. Here, we explore the feasibility and performance of a Convolutional Neural Network (CNN) model using spectrally transformed high-resolution multispectral imagery to map roofprints (i.e., classifying and delineating roofing materials at the building footprint-level) in Washington, District of Columbia (D.C.) and Denver, CO, United States. To generate consistent training data, we integrate geospatial vector data of individual building footprints with real estate industry-derived building-level roofing material data to create labeled image data from Planet SuperDove imagery. We compare the CNN classifier to a pixel-based machine learning (ML) model to demonstrate the capability of our roofprints mapping approach. With F1-scores ranging from 0.56 to 0.95 for the most common roof material classes, the CNN model outperformed the pixel-based ML classifier by 15% and 17% in Washington, D.C., and Denver, respectively. Results demonstrate within-domain robustness for the studied metropolitan areas, which are characterized by differing building densities, roof morphologies, and material patterns. While cross-region transferability was not evaluated, our findings provide a controlled comparison of pixel-based and context-aware approaches for rooftop material mapping and highlight the importance of hierarchical representations that integrate spectral information with roof texture, edge characteristics, spatial arrangement, and neighborhood context for improving classification performance. Accurately mapping building materials has the potential to advance urban planning and environmental policies, including assessments of heat exposure, energy demand, as well as hazard risk and community resilience. Full article