Search Results (22,810)

In our earlier methodology paper, we introduced a hierarchical framework combining graph compression, Diffusion Convolutional Recurrent Neural Networks (DCRNNs), and Multi-Agent Reinforcement Learning (MARL) to approximate Bellman’s optimality principle for real-time energy system control, validated using Palm Springs, California weather data. This extension paper addresses three critical advances. First, we provide a rigorous theoretical proof demonstrating that the DCRNN’s Markovian reduction of strongly path-dependent dynamics yields a bounded approximation error, with the error decaying exponentially in the mixing time of the underlying graph diffusion process. Second, we extend the framework to diverse weather regimes—from stable Mediterranean climates (San Diego) to highly variable marine west coast (Seattle), monsoon (Mumbai), and typhoon-prone regions (Hong Kong)—quantifying how weather-induced path dependence affects the required hidden state dimension and forecasting accuracy. We present comprehensive Leave-One-Out Cross-Validation (LOOCV) results across six cities, demonstrating consistent generalization with R² drops of less than 0.1% under out-of-sample testing. A controlled baseline comparison under matched training protocols shows that the graph-free GRU achieves comparable or higher R² on the one-step prediction task, which we attribute to the near-cumulative structure of the target and the small evaluation graph. We frame the DCRNN’s contribution around its theoretical guarantee and its potential advantage on larger graphs and longer horizons. We also characterize conditions under which the model expects to fail, specifically when weather stochasticity violates the geometric mixing assumption or when the effective temporal correlation length exceeds the GRU’s memory capacity. Third, we outline physics-informed enhancements that are proposed as future development: CFD-integrated loss functions, differentiable Model Predictive Control (MPC) heads, and a modular design enabling alternative turbine configurations. We also propose a standardized rooftop solar thermal deployment architecture with 200 m × 100 m, 100 m × 100 m, and 100 m × 50 m modules designed for data center footprints with pre-allocated HVAC space. We conclude with a stage-gated validation roadmap progressing from unit tests to hardware-in-the-loop simulation to full-scale FEED-site deployment. The completed contributions of this paper are the theorem, its empirical assumption verification, the multi-climate LOOCV study, the matched-protocol baseline comparison, and the sensor-failure robustness analysis. The remaining components are described as proposed extensions. Full article

Background: Gross Berth Productivity (GBP) is a primary terminal benchmark, yet it is routinely read as an efficiency signal, and vessel-call-level evidence is scarce. Methods: From 4463 vessel calls at a major East Asian container terminal (2021–2025), we model GBP with XGBoost and SHAP under a leakage-controlled ablation, adding per-crane, service-level and container-level yard and gate records. Results: Berth productivity falls 24.2%, a medium effect. Gang productivity, the crew-level rate berth productivity commonly read as measuring, falls only 4.2%, a small effect. From 2024 to 2025 berth productivity is statistically equivalent ($\delta =-0.007$), so the decline has leveled off. SHAP identifies cargo volume, not crane speed, as the dominant correlate, stable in 16 of 20 sub-populations and all 200 bootstrap resamples; the model retains $R^2\approx 0.64$ without the ratio’s components. An additive decomposition assigns 71% of the modeled decline to falling call size—fewer containers handled per vessel call, not fewer calls and not smaller ships—and 8% to crane speed; a shift-share decomposition traces 94% of that fall to portfolio turnover. Crane interference is bounded within $\pm 0.10$. Conclusions: Falling berth productivity can reflect a commercial shift to smaller calls rather than operational decline. Evidence is single-terminal. Full article

In the framework of soil erosion processes at different spatial and temporal scales, the concept of connectivity is applied to express the physical link among hillslopes and the channel network concerning flow motion and associated sediments. In this paper, basin sediment connectivity was used in the framework of the Sediment Delivery Distributed (SEDD) model together with sediment yield measurements performed at the W2 experimental basin (Calabria, Italy) in the period 1978–1994. The frequency distribution of the travel time was used to estimate the coefficient β_L of the structural component of the sediment delivery ratio of each morphological unit. Then, using the event sediment yield data, the sediment balance equation was applied to estimate the coefficient β_F of the functional component of the sediment delivery ratio. The analysis demonstrated that the increase in event intensity produces a decrease in β_F and a consequent increase in the sediment delivery ratio. At the annual scale, the sediment balance equation, coupled with a relationship to estimate β_F by the runoff coefficient, allowed for obtaining a good agreement between the measured and calculated sediment yields. The errors in the estimate of basin sediment yield were less than or equal to ±40% for 71.4% of cases. The reliability of the SEDD model is good for both the investigated event and annual scales and improves from the former to the latter, which is, therefore, particularly suited for applications. The model can also be applied to a variety of basins worldwide where it has previously been used without differentiating the structural and the functional connectivity components. Full article

Modern network infrastructures are under attack from increasingly sophisticated attacks that static, rule-based defenses cannot adequately mitigate. We introduce a multilayer anomaly detection and mitigation framework for Software-Defined Networking (SDN) environments consisting of four integrated subsystems: (i) a Micro-segmentation Integrated Management and Defense System (MIMDS), (ii) an Adaptive CNN-LSTM-Attention Deep Packet Inspection (MCLA-DPI) module, (iii) a Hybrid Adaptive Cyberattack Prediction (KBGM) framework, and (iv) an AI-driven log analysis pipeline. Detection, prediction and containment operate in parallel (as opposed to conventional approaches) and correlate results through a common risk-scoring mechanism to coordinate policy enforcement via OpenFlow and P4-compatible data planes. The experimental evaluation shows promising performance. MIMDS achieves 94.3%. detection accuracy with full traffic isolation in 10.1 s. MCLA-DPI achieves 98.9% classification accuracy with 14 ms inference latency, outperforming baseline models SVM and LSTM on encrypted traffic. KBGM achieves 98.4% detection accuracy with 7.2 ms mean response time on the CICIDS2017 dataset. All modules are trained in federated learning to preserve data locality with continuous improvements of the global model. The results collectively demonstrate quantifiable improvements over single-paradigm approaches in detection fidelity, response latency, resource efficiency, and privacy compliance. An ablation study isolates the contribution of integration itself. Removing the coordination layer while retaining all four detectors reduces accuracy from 0.892 to 0.634 and raises the false-positive rate from 0.031 to 0.436, while peak rule installation rises from 22.3 to 98.4 rules per second and oscillation events increase by two orders of magnitude; the integrated framework also exceeds its strongest individual subsystem, which reaches 0.831 accuracy at a false-positive rate of 0.117. These figures are obtained from the released reference implementation over a synthetic campaign and are reported separately from the component measurements. Full article

Background/Objectives: Periodontitis, the sixth most prevalent disease worldwide, affects over 740 million people and disproportionately burdens transitional economies. Although artificial intelligence (AI)—including deep learning (DL) and machine learning (ML)—achieves high diagnostic accuracy in research settings, a gap persists between proof-of-concept and real-world deployment, especially where regulation is nascent, as in Kazakhstan. This scoping review maps global evidence on AI for periodontal diagnosis, risk prediction, and monitoring; evaluates governance frameworks; and proposes a contextualised implementation model for emerging health systems. Methods: Following PRISMA-ScR and PRISMA 2020 guidance, PubMed/MEDLINE, Scopus, Web of Science, Embase, and Cochrane Library were searched (January 2015–April 2025), supplemented by regulatory and grey literature; ten additional sources published after the search closure were subsequently identified through citation checking and expert peer review during revision, as a targeted amendment rather than a re-executed database search. Forty sources in total (29 peer-reviewed empirical and review studies plus 11 regulatory, grey literature, and patent documents) met the inclusion criteria and were charted thematically. Results: Two dominant paradigms emerged: image-based DL (convolutional neural networks and Vision Transformers), achieving 73–98.6% accuracy for radiographic bone loss detection, and ML-based non-clinical screening using patient-reported data and salivary biomarkers. Digital tools (smart toothbrushes, chatbots, and IoT platforms) form a third domain. Performance dropped consistently on external validation, reflecting data quality and sample size constraints. Regulatory analysis showed convergence of the EU AI Act, U.S. FDA framework, WHO guidance, and Kazakhstan’s AI Development Concept (2024–2029) around risk-based classification, transparency, and post-market surveillance. Conclusions: Safe, effective AI integration in periodontology requires a phased approach: national multimodal databases, local clinical validation, certified workflow integration, continuous monitoring, population-level surveillance, and legal governance covering liability and insurance. Kazakhstan’s evolving regulatory and digitalisation strategy may offer a context-specific case for evaluating AI adoption pathways across Central Asia and transitional economies, pending prospective validation. Full article

Measuring bite force is essential for assessing the masticatory system and diagnosing oral disease. Existing measurement devices have low spatial resolution and susceptibility to electromagnetic interference. This paper presents a machine learning (ML)-assisted distributed fiber optic sensing system based on Scattering Level Multiplexing (SLMux) for high-resolution bite force analysis. Enhanced backscattered data were acquired through optical backscattered reflectometry from 88 sensing points along the dental arch. Measured data were reconstructed into a two-dimensional map of bite force and analyzed through an ML pipeline. Sector classification across 4 regions and weight prediction were processed by an end-to-end fine-tuned ResNet-18 Convolutional Neural Network (CNN) and classical ML approaches. ResNet-18 is compared with Logistic Regression, Support Vector Machine (SVM), Random Forest, XGBoost (Extreme Gradient Boosting), Extra Trees, and k-Nearest Neighbors (kNN) trained on handcrafted features. On sector classification, Logistic Regression achieved the best performance (98.71% accuracy). On the weight prediction task, formulated as a 12-class problem, the end-to-end ResNet-18 CNN substantially outperformed all classical models, reaching 51.28% accuracy and a mean absolute error of 78 g, versus 124 g for the best classical model. A regression-based ResNet-18 variant was also trained on the wavelength-shift and weight data, resulting in a mean absolute error of 62.6 g. The results indicate that integrating ML with distributed fiber-optic sensing has the potential to enhance dental diagnostics and treatment planning. Full article

Advanced Metering Infrastructure (AMI) underpins bidirectional communication in smart grids, a foundational layer of smart city energy systems, facilitating the flow of data, real-time monitoring, and demand-responsive control. But its connectivity exposes smart meters to data tampering, denial-of-service attack, and false-data-injection attacks. Most intrusion detection systems (IDSs) for AMI only report that an intrusion has occurred but cannot reconstruct how it propagated or where it originated, leaving the operators without the forensic evidence to perform containment. This paper proposes a multi-stage approach coupling detection with forensic analysis. A recurrent neural network (RNN) extracts temporal features, a support vector classifier (SVC) performs binary classification, and ant colony optimization (ACO) serves two purposes: feature selection before classification and a backward path reconstruction after an intrusion is confirmed. The proposed framework is evaluated on a simulated AMI network with forensic ground truth and further validated on the public UNSW-NB15 benchmark. The detection accuracy exceeds $96\%$, while ACO reduces the feature set from 40 to 14. A McNemar’s test ($p=0.265$) indicates that this feature reduction does not significantly alter the per-sample error pattern. With the use of the improved tracer, the Path Overlap Score increases from $0.29$ to $0.40$, while the False-Positive Path Rate decreases from $0.39$ to $0.19$, relative to the centroid baseline tracer used for forensic tracking. This improvement in the Path Overlap Score is statistically significant ($p=4.39\times {10}^{-8}$). However, the Source Localization Rate remains relatively low ($9\%$–$11\%$) for both methods, owing to the intrinsic difficulty of identifying the true source meter from incomplete alert data. Therefore, the proposed framework not only reliably detects intrusions but also significantly outperforms the baseline tracer in path overlap and false-positive rate, although precise source localization remains an open challenge. Full article

Parent participation is mandated in special education and transition planning for students. However, immigrant Filipino parents face numerous challenges related to immigration, navigating special education, and transitions. To understand their participation in their children’s education, it is important to understand their history, immigration experiences, and beliefs. This study explores how immigrant Filipino parents advocated for their children in New York City’s public schools and how their immigration experience has shaped their participation. This qualitative research study conducted in-depth, semi-structured interviews with two Filipina immigrant mothers. Data were analyzed using thematic analysis informed by Bourdieu’s theory of capital. Two themes were identified from the data: (1) uncovering the mothers’ habitus through their ways of participation and use of capital and (2) communities and social networks as supports for parent participation and child-reading. The mothers possess capital, such as language, which is useful in accessing supports and services for their children. Their values and beliefs regarding education, disability, and their role as parents are shaped by Filipino values, which are also influenced by years of colonial influence on the Philippines. Educators and schools should be proactive in coordinating efforts of parents and families as they plan for the student’s education and transition. Full article

Accurate plant height estimation is important for monitoring crop growth and supporting precision agricultural management. Manual measurements are labor-intensive, while LiDAR-based methods are expensive and require complex processing. UAV photogrammetry provides a lower-cost alternative but remains challenging in flooded rice paddies because of canopy deformation and difficulties in terrain extraction. This study proposes Rice-STNet, a spatiotemporal deep learning framework for end-to-end rice plant height estimation using multi-temporal UAV RGB imagery. Rice-STNet integrates a convolutional neural network for spatial feature extraction, Time2Vec for temporal encoding, and a gated recurrent unit network for modeling temporal dependencies across observation dates. The framework was evaluated using field data collected from rice paddies over two growing seasons. Rice-STNet achieved an R² of 0.97, a root mean squared error of 1.97 cm, and a mean absolute error of 1.14 cm. It outperformed random forest, support vector regression, a CNN-only baseline, and a UAV photogrammetry-based point-cloud approach. In addition, the framework generated high-resolution plant height maps for field-scale analysis of spatial growth variability. These results underscore the importance of jointly modeling spatial and temporal characteristics for continuously evolving crop traits. The proposed framework offers an accurate, scalable, and non-destructive solution for large-scale crop phenotyping and precision agriculture. Full article

Background: A woman’s childbirth experience is important to her health and family well-being. Evidence-based intrapartum care may support a positive experience. When resources are constrained, identifying services that are most strongly associated with childbirth experience may help inform quality-improvement priorities. This study aimed to explore priority intrapartum services associated with women’s experience following vaginal birth. Methods: A multicenter cross-sectional study was conducted in mainland China. A total of 498 postpartum women from ten tertiary facilities were involved. Data were collected through retrospective review of medical records and self-administered questionnaires completed by postpartum women. A self-designed questionnaire was used to collect general information and information on intrapartum services, and the Childbirth Experience Questionnaire was used to assess childbirth experiences. Random forest and Bayesian network analysis were used to examine associations and identify potentially high-priority services. Results: Childbirth experience scores ranged from 27 to 76, with a mean of 61.29. In bootstrap analyses of the random forest model, coefficient of variation for variable-importance estimates ranged from 5.1% to 32.3%. The Bayesian network contained seven structurally compact communities, five of which were closely associated with childbirth-experience outcomes; hospital type was an important connecting node between intrapartum services and childbirth experience. Among the top ten factors associated with the overall childbirth experience, several intrapartum services stood out. These included the type, initiation time, and duration of skin-to-skin contact; the number of evidence-based services received during the first and second stages of labour; and the use of pain-relief methods. Conclusions: Receiving a broader range of evidence-based intrapartum services, longer uninterrupted skin-to-skin contact and access to pain-relief options were associated with more positive childbirth experience scores. Full article

Energy conservation remains a fundamental challenge in wireless sensor networks (WSNs), where battery-powered nodes are often deployed in locations where recharging is impractical. Because spatially correlated sensors frequently report redundant readings, deactivating less informative sensors can extend network lifetime while retaining nearly the full-network classification performance. Prior sensor-selection studies, however, have often evaluated fixed sensor counts on public classification benchmarks without ground-truth labels identifying which sensors are genuinely informative, while estimating energy savings primarily from sensor-count ratios rather than topology-dependent physical energy models. This paper presents an adaptive sensor-selection framework that uses gradient-boosted trees with permutation importance to rank sensors and identify the smallest subset that retains a predefined fraction of full-network classification accuracy. The framework is evaluated using a purpose-built synthetic benchmark with known ground-truth sensor informativeness and an explicit network topology governed by a first-order radio energy model, together with eight independently re-implemented baseline methods evaluated under identical conditions across eight data-partition seeds. At the selected operating point, GBM-Permutation retains 99.3% of the full-sensor classification accuracy (84.6% versus 85.2%) using only 10 of 54 sensors, corresponding to an 82.6% reduction in modeled per-round energy consumption; modeled first-node-dies network lifetime increases from 615 rounds (using all 54 sensors) to 883 rounds (a 1.4-fold increase), distinct from the sensor-count-based Lifetime Extension Factor (LEF) of 5.4. The proposed method significantly outperforms all baseline methods in the primary synthetic evaluation (p < 0.01). A redundancy–severity sensitivity analysis and validation on two generic real-world tabular datasets and one real sensor-derived dataset show that the advantage is strongest under moderate redundancy rather than being universal. These findings demonstrate the value of ground-truth-validated and topology-aware evaluation for identifying sensor-selection methods that are effective under energy-constrained WSN conditions. Full article

Due to the limitations of seismic data resolution, conventional inversion methods are commonly limited by the band-limited nature of seismic data, noise interference, and insufficient resolution for subtle geological targets. This study proposes a continuous wavelet transform (CWT)-assisted multiscale stepwise learning (MSL) workflow for seismic impedance inversion. The proposed workflow transforms seismic traces into time–frequency representations using CWT and progressively learns different frequency components through convolutional modules with different kernel sizes. A seismic forward-modeling constraint is further introduced to improve the consistency between the predicted impedance and the observed seismic response at non-well locations. The network takes a single seismic trace represented by CWT coefficients in the 1–80 Hz frequency range and a corresponding low-frequency initial impedance model as inputs, and outputs the acoustic-impedance trace at each time sample. The main methodological novelty is the progressive learning of different frequency components using frequency-dependent convolutional kernels. The method is evaluated on a 3D seismic dataset from the Sichuan Basin, China, using leave-one-well-out validation involving seven wells. In addition, we provide detailed ablation experiments and quantitative analysis. Across all seven blind-well tests, the proposed method consistently achieves the highest mean correlation coefficient (0.8509) and the lowest mean MSE (0.0636) among the compared methods. These results indicate that the proposed workflow can effectively integrate multiscale seismic information and provide effective support for seismic impedance inversion. Full article

Objectives: Cardiovascular diseases (CVDs) represent a major public health challenge in Thailand, but consumer preferences for preventive nutraceuticals remain poorly understood. This study aimed to assess preferences and willingness to pay for nutraceuticals for cardiovascular disease prevention among the Thai population. Methods: A discrete choice experiment (DCE) was undertaken with five attributes, including clarity of evidence, effectiveness, safety, cost, and recommended information sources identified by a comprehensive literature review and qualitative consumer interviews. Thai citizens aged 20 years or older with no history of CVD and able to understand Thai were eligible for this survey. Data collection was conducted using an online survey distributed via social networks between March and April 2026, and the data were analyzed using random parameters logit and latent class logit models. Results: A total of validated responses from 732 Thai adults were used. Clarity of evidence was the most influential attribute (relative importance: 38.3%), followed by recommendation source (24.3%), effectiveness (15.2%), safety (14.2%), and cost (7.9%). High-quality scientific evidence and healthcare professional recommendations were strongly preferred, yielding the highest monthly WTP of 2215.94 THB (US$67.06) and 2080.28 THB (US$62.95), respectively. Recommendations from social media influencers were viewed unfavorably. LCL analysis identified three distinct consumer segments as value-conscious (37.1%), evidence-oriented (28.4%), and healthcare professional-oriented (34.4%). Conclusions: Thai consumers prioritize scientific evidence and healthcare professional guidance over social media or commercial marketing when choosing cardiovascular nutraceuticals. Effective communication and product labeling strategies should be tailored to these distinct consumer segments to support evidence-based self-care. Full article

Urban disaster screening requires transparent methods that can integrate heterogeneous open datasets without implying unsupported causal or probabilistic relationships. This study presents an open-data framework for screening flood-footprint population proxies and potential disruption to hospital accessibility in Bucharest, Romania. The analysis covers 104 archived hexagonal spatial units and combines a 100-year flood-depth scenario from the Joint Research Centre, WorldPop 2020 population estimates, hospital-routing outputs derived from OpenStreetMap, and a publicly available seismic screening surface from the European Facilities for Earthquake Hazard and Risk. Flood and seismic information are retained as distinct screening dimensions because the available data do not support modelling their causal interaction, temporal sequence, joint probability, earthquake-related infrastructure damage, or hospital capacity. For spatial units that remain connected to a hospital, a transparent two-domain service-priority index is calculated using flood-footprint population proxy and the potential change in hospital accessibility under the flood scenario. Units for which no hospital route is available are reported separately as binary service-disconnection alerts rather than being assigned an arbitrary numerical penalty. The robustness and interpretability of the framework are examined through network monotonicity, score boundedness, Pareto dominance, penalty-free rank invariance, and rank-acceptability analysis. The results are communicated using a Pareto frontier, rank trajectories across the full weight simplex, indicator-contribution decomposition, and an exact hypergeometric assessment of class overlap. The proposed framework provides a transparent first-order planning tool for identifying locations that may require more detailed investigation. It should be interpreted as a screening approach rather than as a probabilistic risk, cascading-hazard, infrastructure-damage, or service-loss model. Full article

Background: Osteoporosis (OP) arises from dysregulated bone metabolism driven by genetic and epigenetic factors. Osteoblasts are central to bone formation, and their functional heterogeneity—shaped by genetic background and receptor expression profiles—may critically influence OP susceptibility. This study aimed to identify osteoblast-specific genes robustly associated with OP and elucidate their underlying pathogenic mechanisms. Methods: We analyzed single-cell RNA sequencing (scRNA-seq), bulk RNA sequencing (bulk RNA-seq), and spatial transcriptomics (ST) datasets of osteoblasts. Differentially expressed genes (DEGs) were identified from scRNA-seq and bulk RNA-seq datasets, followed by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis. Three machine learning methods and an artificial neural network (ANN)-based weighting analysis, together with weighted gene co-expression network analysis (WGCNA), were used to prioritize candidate OP-associated genes. ST data were integrated with CellChat analysis based on scRNA-seq data to investigate spatial expression patterns and potential cell–cell communication. Candidate genes were validated by immunohistochemical staining in human femoral head samples and quantitative real-time PCR (qRT-PCR) in MC3T3-E1 cells. Results: The intersecting DEGs from the scRNA-seq and bulk RNA-seq datasets were potentially related to inhibition of ossification. Three machine learning methods identified five osteoporosis-associated candidate genes: TTYH3, NRBP2, MPG, HSPG2, and GPR153, which were subsequently evaluated using ANN-based weighting analysis. Among these, TTYH3 and MPG were identified as OP-related genes by WGCNA. Integration of the scRNA-seq, ST, and CellChat results suggested that SPP1 was highly expressed in osteoblasts from the OP sample and exhibited a spatially heterogeneous expression pattern. Immunohistochemical staining of human femoral head tissues from individuals with normal bone mass and osteoporosis, together with qRT-PCR analysis in MC3T3-E1 cells, further validated the differential expression of TTYH3 and MPG. Conclusions: This study identified candidate osteoporosis-associated biomarkers and provided insights into potential pathogenic mechanisms, thereby establishing a basis for future mechanistic and clinical validation. Full article