Search Results (2,776)

Accurate cryptocurrency price forecasting represents an important yet challenging problem in financial time-series analysis due to the highly volatile, nonlinear, and noise-sensitive nature of digital asset markets. Although transformer-based architectures have recently demonstrated strong capabilities in temporal sequence modeling, their behavior under high-frequency cryptocurrency dynamics and the role of explainability-guided feature refinement remain insufficiently explored. To address this gap, this study presents a comprehensive transformer-based forecasting framework for hourly cryptocurrency price prediction and investigates the impact of explainability-guided feature optimization on forecasting performance, robustness, and interpretability. Five transformer architectures—Vanilla Transformer, Informer, Autoformer, Reformer, and Temporal Fusion Transformer (TFT)—are systematically evaluated across five major cryptocurrency assets: Bitcoin (BTC), Ethereum (ETH), Solana (SOL), Dogecoin (DOGE), and Ripple (XRP). The experimental framework employs Open, High, Low, Close, and Volume (OHLCV) data together with a broad set of engineered technical indicators and evaluates model performance using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). To improve interpretability and reduce feature redundancy, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) are integrated directly into the forecasting pipeline. Based on the resulting explanations, asset-specific feature subsets are constructed, and all models are subsequently retrained using the refined feature representations. The results show that explainability-guided feature refinement provides compact, model-aware, and interpretable feature subsets with competitive forecasting performance; however, its effect on prediction accuracy is dependent on the cryptocurrency asset, transformer architecture, retained feature subset size, and market conditions. Additional robustness, sensitivity, alternative feature-selection, and statistical significance analyses indicate that the SHAP–LIME Top-15 subset should be interpreted as a conservative dimensionality-reduction strategy rather than a universally optimal feature-selection rule. The findings further reveal that transformer architectures incorporating sparse attention, decomposition mechanisms, or gating structures generally provide stronger performance than the Vanilla Transformer under highly volatile hourly market conditions. Overall, the proposed framework demonstrates that combining transformer-based forecasting with explainability-guided feature refinement can support interpretable and parsimonious high-frequency financial time-series modeling, while highlighting the importance of evaluating robustness, feature-selection sensitivity, and statistical variability alongside average forecasting errors. Full article

Accurately forecasting carbon emission trends in China’s power sector is of great significance for achieving the “dual-carbon” goals, optimizing the energy structure, and formulating low-carbon development strategies. This paper aims to develop a forecasting method capable of effectively characterizing the long-term evolutionary patterns and short-term dynamic features of carbon emissions in China’s power sector. However, existing models struggle to simultaneously describe the nonlinear S-shaped growth trend, periodic fluctuations, and long-term memory effects inherent in carbon emission time series. To address these issues, this paper proposes a rational canonical form grey Gompertz forecasting model based on the Hausdorff fractional derivative. By introducing a rational canonical form matrix structure, this model enhances the capability of the grey Gompertz model to represent multi-scale periodic information, and incorporates the Hausdorff fractional derivative to characterize the non-local dynamic features during the time-series evolution, thereby improving the model’s adaptability to complex nonlinear carbon emission sequences. Taking the quarterly and semi-annual carbon emission data of China’s power sector as the research object, simulation and forecasting experiments under various time scales and sample settings were conducted to verify the effectiveness of the new model, and a comparative analysis was performed against traditional grey models, fractional-order grey models, and statistical forecasting models. The results indicate that the model possesses certain advantages in structural representation and dynamic memory mechanisms, exhibiting high forecasting accuracy and stability across different time scales. The full-sample mean absolute percentage errors (MAPEs) were all below 3%, and the MAPEs of the optimal schemes for quarterly and semi-annual data reached 0.8353% and 0.0491%, respectively. Finally, the model was utilized to effectively forecast the carbon emissions of China’s power sector for the 2026–2027 period. Full article

The running-in phase of hermetic reciprocating compressors is characterized by slow, nonstationary changes in operational signals and is commonly assessed using empirically defined test durations or manually defined indicators. From a data-analysis perspective, this constitutes a feature extraction problem, as informative temporal patterns might be identifiable from long and noisy time series acquired during operation. This paper investigates the use of Long Short-Term Memory (LSTM) networks as a data-driven feature extraction mechanism for running-in analysis based on electrical current and vibration measurements. LSTM-based sequence encoding is used to transform raw time-series segments into compact representations that summarize their temporal structure and enable discrimination between running-in and steady-state regimes, as well as characterization of intermediate conditions in the learned feature space using a multilayer perceptron. Model structure, input configuration, and segmentation parameters are selected through an optimization procedure under weak supervision. Experimental results show that the extracted LSTM features capture consistent temporal signatures of running-in across different compressor units, with one unit exhibiting outlier behavior in latent space analysis. The results support the use of recurrent neural networks for noninvasive running-in assessment in rotating machinery and related condition monitoring tasks. Full article

Background and Aim: Studies worldwide have documented a reduction in acute myocardial infarction (AMI) admissions during the COVID-19 pandemic. We aimed to assess AMI admissions, management, and in-hospital outcomes in Kosovo during the pandemic period. Methods: We conducted a retrospective single-centre study including all consecutive AMI admissions from 1 January 2018 to 31 December 2020 at the Clinic of Cardiology, University Clinical Centre of Kosovo, the only public tertiary primary percutaneous coronary intervention (PCI) centre in the country. The pre-pandemic (PP) period was defined as 1 January 2018–12 March 2020 and the ongoing pandemic (OP) period as 13 March–31 December 2020. Demographic, clinical, laboratory, angiographic, and outcome data were collected. In-hospital mortality was defined as death between admission and discharge. Interrupted time-series analyses were performed. Results: A total of 898 AMI patients were admitted during OP and 2955 during PP. Daily admissions decreased from 3.8/day to 3.05/day (p < 0.001). OP patients were younger, more often hypertensive and smokers, and more frequently male. Coronary angiography (81% vs. 76%, p = 0.001) and PCI (66% vs. 60%, p = 0.001) were more frequent during OP, while in-hospital mortality remained unchanged. Time-series analysis showed a sharp fall in admissions at OP (β = −12.66, p < 0.001) with gradual recovery, alongside an increase in PCI use (β = 0.184, p < 0.001) and stable mortality. The proportion of ST-segment elevation (STEMI) cases increased significantly during the pandemic period. Compared with the pandemic period, admission during the pre-pandemic period was associated with lower odds of undergoing PCI (aOR 0.81, 95% CI 0.69–0.95), while age ≥ 65 reduced the likelihood of PCI. Conclusions: AMI hospitalizations in Kosovo declined significantly during the COVID-19 pandemic. The pandemic cohort was younger, with the profile of admitted patients shifting toward younger age, male sex, and STEMI presentation. Despite fewer admissions, PCI use increased, while no statistically significant difference in in-hospital mortality was observed between the two periods. These findings describe changes in treatment patterns and short-term in-hospital outcomes during the pandemic but should not be interpreted as evidence of equivalent overall quality of AMI care. Full article

This paper presents a significant evolution of the Identification of Active Anomalies (IAA) system, moving beyond previous descriptive frameworks by integrating an advanced machine learning pipeline for automated, real-time Structural Health Monitoring (SHM). Utilizing acoustic emission (AE), the upgraded IAA framework combines signal clustering, image recognition, and machine learning to monitor the structural condition of a highway overpass located near a major urban agglomeration. The monitoring results provide a reliable foundation for assessing structural health and implementing automated traffic control, which is essential to ensure safe operations. Unlike baseline implementations, this intelligent system extracts multi-parametric features using Principal Component Analysis (PCA) and transforms temporal wave streams into Continuous Wavelet Transform (CWT) scalograms. These visual representations are processed by a custom 14-layer Deep Convolutional Neural Network (CNN) combined with an unsupervised Self-Organizing Map (SOM) to eliminate operational noise and classify internal failures. AE signals recorded under service loads undergo multi-parametric analysis using pattern recognition techniques and are assigned to specific classes corresponding to active anomalies within the material or structure. Each class is linked to a distinct structural hazard level, ranging from safe operation to a critical loss of structural safety. Corresponding traffic control measures, including vehicle speed and weight restrictions, are dynamically introduced to maintain operational safety. To validate the scalability of the framework, this study synthesizes statistical data across a comprehensive fleet of 180 monitored bridge structures, backed by a predictive ARIMA time-series model that forecasts residual service life. The proposed methodology was experimentally validated on an A2 highway overpass, a vital component of the Łódź transport hub that facilitates north–south and east–west transit in Poland. The IAA system functions as a proactive diagnostic tool for infrastructure management agencies, preventing sudden, unforeseen structural failures. Ultimately, it enables the efficient and safe operation of a Smart City while ensuring that maintenance funds are rationally and optimally allocated. Full article

Generative artificial intelligence (GenAI) has gained prominence in scholarly communication, yet its lexical classification, disciplinary diffusion, semantic organization, and association with citation-based recognition remain understudied. This study analyzes 487,753 research and review articles from OpenAlex (August 2019–March 2026), combining rule-based lexical classification with interrupted time-series models, disciplinary mapping, keyword co-occurrence analysis, and citation-count models. The results show a discrete level change in the visibility of GenAI terminology after November 2022. This discontinuity remained robust under classification-error sensitivity analyses, whereas inference regarding the subsequent slope was more sensitive to classification assumptions. The proportion of GenAI-classified publications captured by high-specificity lexical rules increased from 86.65% before December 2022 to 93.64% afterward, consistent with greater lexical concentration. GenAI classification was broadly but unevenly distributed across fields, with the highest prevalence in Computer Science (49.42%) and Decision Sciences (38.57%). A curated keyword network showed overlapping thematic concentrations rather than sharply separated semantic subfields. Month-adjusted citation models showed a positive association between high-specificity classification and citation counts, with a smaller association with GenAI citations after December 2022. Overall, the study characterizes the temporal, lexical, disciplinary, thematic-network, and citation-related dimensions of GenAI’s emergence in scholarly communication. Full article

Large-scale ecological restoration has greatly boosted carbon sequestration across China’s Loess Plateau, yet the long-term sustainability of restoration-fueled carbon growth remains unclear. This study constructs a dynamic carbon stock mapping framework integrating GEDI LiDAR, Landsat time-series data and the dynamic InVEST model for ecosystem carbon accounting to analyze carbon accumulation trends from 2000 to 2025. Regional total carbon storage rose from 8088.65 Tg C to 9269.54 Tg C, with a net gain of 1180.89 Tg C, but carbon growth slowed markedly. The share of regions with notable carbon growth dropped sharply from 77.92% (2000–2013) to merely 1.50% (2013–2025). Logistic modeling shows 2025 carbon storage hit roughly 96.3% of its estimated saturation threshold, meaning room for fast carbon accumulation is shrinking. SHAP analysis identifies SWIR1 (31.1%) and NDVI (15.9%) as primary predictors contributing to biomass carbon variation, with nonlinear relationships proving moisture and vegetation jointly shape carbon accumulation. PLUS model simulations show ecological priority land-use scenarios deliver stronger carbon storage capacity than cropland protection or natural development schemes, though future carbon increments will be far smaller than historical gains. Ultimately, this study confirms that the Loess Plateau carbon sink is transitioning from restoration-facilitated rapid expansion to environmentally constrained slow growth, with current carbon storage approaching 96.3% of the regional ecohydrological carrying capacity. Full article

Industrial fault detection and diagnosis are essential for maintaining operational reliability and minimizing performance degradation in process industries. This study presents an integrated machine learning framework for fault detection, fault-type classification, and fault severity estimation using a synthetic chemical-process time-series dataset comprising six reactors operating under multiple conditions. The framework combines a five-class fault diagnosis model with a gated severity estimation stage that is activated only when a fault is detected, enabling simultaneous assessment of process condition and operational impact. Eight process variables were selected through statistical and process-oriented analysis, while one-minute difference features and reactor identity information were incorporated to capture short-term process dynamics and equipment-specific operating characteristics. The framework was evaluated using an episode-aware methodology incorporating fault-episode partitioning, leakage-prevention measures, grouped cross-validation, and episode-level analysis. The selected classification model achieved a balanced accuracy of 76.25% and a macro F1-score of 82.44%. For severity estimation, the complete end-to-end pipeline achieved R² = 0.153 across active-fault observations, illustrating the impact of fault detection errors on downstream severity assessment. When evaluated across all observations, including predominantly normal conditions, the corresponding R² increased to 0.871. Under an oracle scenario using the true fault type, severity estimation achieved R² = 0.920. The results provide fault-specific and episode-level insights and support a proof of concept within this synthetic industrial process environment. Full article

The North China Plain (NCP) has entered a “New Water Regime” driven by water diversions, stringent groundwater management, and climatic fluctuations, resulting in the widespread recovery of regional groundwater levels. How land subsidence has responded spatiotemporally to these interventions remains unclear. To quantify the regional hydrogeological responses, we integrated Time-Series Interferometric Synthetic Aperture Radar (TS-InSAR) observations spanning 2016–2023 with weighted centroid tracking and Emerging Hot Spot Analysis (EHA) based on a Space-Time Cube. Centroid tracking indicated that the regional subsidence center remained spatially stable without significant directional drift. In the western and southern regions, Persistent and Intensifying Cold Spots dominated, corresponding to ground stabilization and rebound. In the central-eastern centers, 26.80% and 6.47% of the statistically significant clustering area were Persistent and Intensifying Hot Spots, reflecting continuous localized compaction, with Mann–Kendall and Sen’s slope analyses indicating a decelerating trend within these clusters. A further 4.92% and 9.63% were Diminishing and Historical Hot Spots, indicating reduced clustering after 2019. Overall, subsidence has decelerated markedly across most of the plain, yet compaction inertia in thick clay layers sustains localized hot spots despite rebounding groundwater levels. This time-lag response implies that short-term water-level recovery cannot immediately reverse subsidence, necessitating sustained groundwater management across the NCP. Full article

It remains unclear why deformation results obtained by applying different MT-InSAR methods to the same dataset, as well as those from different SAR datasets, exhibit discrepancies. Hence, this study selects Baode County, located in the Loess Plateau, as the study area to conduct a comparative analysis of multi-source SAR data and multiple MT-InSAR techniques for surface deformation monitoring. The datasets consist of concurrent Radarsat-2 and Sentinel-1 images acquired from October 2020 to June 2024. PS-InSAR, SBAS-InSAR, and IPTA-InSAR are adopted to compare their applicability across multiple dimensions, such as point coverage, deformation correlation, and mapping performance. Furthermore, based on IPTA-InSAR, the influences of spatiotemporal resolutions from different datasets on monitoring results are analyzed. The Sequential Turning Point Detection (STPD) method is incorporated to characterize the dynamic evolution of deformation and its correlation with precipitation. The results indicate that the three techniques exhibit favorable consistency and complementarity across diverse landform types. Nevertheless, SBAS-InSAR’s superiority in point density cannot be translated into an ability to represent continuous deformation fields. In contrast, IPTA-InSAR demonstrates better adaptability to complex surface conditions. Regarding data sources, Radarsat-2 achieves superior monitoring performance thanks to its high spatial resolution, yet its monitoring capacity is sensitive to variations in spatial resolution. Multi-looking not only reduces the maximum subsidence rate by approximately half but also increases elevation uncertainty by up to 24.8% and deformation-rate uncertainty by up to 42.1%. Sentinel-1, with its shorter revisit cycle, provides temporal sampling advantages that significantly enhance the ability to capture rapid subsidence signals. Through controlled-variable experiments, a dual-comparison analysis of multi-source SAR datasets and multiple MT-InSAR techniques enables the separation of discrepancies induced by algorithms from those originating in the datasets. The multifaceted experimental design provides a relatively comprehensive assessment of the influencing factors. The findings of this study can serve as references for InSAR data source combinations, technique selection, results presentation, and reliability assessment of deformation results in complex monitoring areas. Full article

Financial markets exhibit complex nonlinear dynamics driven by interactions between firm-specific characteristics and macroeconomic conditions, requiring robust mathematical models capable of capturing structural changes and temporal heterogeneity. This study proposes an explainable machine learning framework that combines regime-switching analysis and predictive modeling to investigate the long-term resilience of firms listed in the BIST 100 Index. Structural breaks in monthly return and volatility series are first detected using the Pruned Exact Linear Time (PELT) algorithm, enabling the classification of firm trajectories into resilient and fragile market regimes. A Regime Persistence Score is then introduced to quantify the proportion of time each firm remains in the resilient state. This score serves as the response variable in a Random Forest model that evaluates the influence of firm-specific financial indicators and macroeconomic variables, including exchange rates, producer price inflation, and commercial loan interest rates. In addition, a forward-looking classification model estimates the probability that firms will transition into a fragile regime within the subsequent six months. The proposed framework achieves an AUC of 0.706 and demonstrates stable predictive performance under alternative regime definitions, penalty parameters, cost functions, and cross-validation strategies. Explainability analysis derived from Shapley Additive Explanations (SHAP) data shows book-to-market ratio and sensitivities to inflation and interest rate changes as the most important components of enduring resilience. The proposed methodology provides an interpretable mathematical framework for regime detection, nonlinear time-series modeling, and decision support in sustainable financial systems, offering practical value for risk assessment and resilience-oriented portfolio management. Full article

Background: Inappropriate antibiotic use contributes to antimicrobial resistance. Evidence regarding the impact of antimicrobial stewardship programs (ASPs) in pediatric intensive care units (PICUs) remains limited. The study aims to evaluate the impact of a multidisciplinary prospective post-prescription review and feedback (PPRF) ASP in a European PICU on antimicrobial use and prescription quality (PQ). Methods: A prospective quasi-experimental study was conducted using interrupted time-series analysis to evaluate antimicrobial use trends before (July 2018–December 2020) and after (January 2021–March 2025) the implementation of a multifaceted ASP, integrated within a hospital-wide ASP, in a tertiary 24-bed medical–surgical PICU in Barcelona, Spain. Antimicrobial use was measured as days of therapy per 100 patient-days (DOT/100 PD) and per 100 discharges (DOT/100 D) and was analyzed by WHO AWaRe group and by individual drug. PQ was evaluated by means of cross-sectional point-prevalence surveys (PPSs). Results: During the study-period, median monthly activity in the PICU remained stable: 475 patient-days (IQR: 396–534) and 113 patients discharged (IQR: 102–124). Following ASP implementation, total antibiotic consumption significantly decreased, with an immediate reduction of 18.2 DOT/100 patient-days (p = 0.006) and a sustained monthly decline of 0.73 DOT/100 patient-days (p = 0.028). The COVID-19 pandemic did not significantly affect overall antibiotic consumption. Access and Reserve antibiotic use decreased significantly after ASP implementation, whereas Watch antibiotic use remained unchanged. Overall antifungal consumption was not significantly modified, although a transient increase was observed during the pandemic. At the individual drug level, several antibiotics demonstrated significant immediate decreases after ASP implementation. Prescription quality remained high (>88% optimal prescriptions), and PICU length of stay and mortality were unchanged. Conclusions: Implementation of a multidisciplinary PPRF-based ASP in a tertiary PICU was associated with a sustained reduction in antibiotic exposure without compromising prescription quality or clinical safety. These data support the effectiveness of collaborative stewardship in one of the most complex pediatric healthcare settings. Full article

Advance prediction and management of ground subsidence are crucial, as its occurrence can lead to human casualties and property damage, particularly in densely populated metropolitan areas. This study developed artificial intelligence (AI)-based models to predict the daily occurrence of urban ground subsidence using meteorological factors. Focusing on selected areas within South Korea, a daily time-series dataset spanning 2010–2015, the primary analysis period selected for record consistency, was constructed using daily precipitation, temperature, and ground subsidence occurrence records. The predictive performance of seasonality-based baselines, conventional machine-learning models (random forest, extreme gradient boosting (XGBoost)) and deep-learning models (long short-term memory (LSTM), LSTM-Transformer (LT)) was evaluated under a strictly chronological, leakage-free protocol with multi-seed repetition and bootstrap confidence intervals. Antecedent meteorological conditions provided predictive skill significantly beyond seasonal climatology; notably, this skill was captured most effectively by a logistic regression on a compact summary of the preceding day’s conditions (macro F1 = 0.608, ROC-AUC = 0.655), which the deep sequence models matched but did not exceed. Analyses of input sequence length showed that short windows outperformed longer ones, and temperature variables emerged as the dominant predictors, indicating that recent antecedent conditions—rather than extended meteorological sequences or model complexity—carry most of the predictive information. This study confirms the feasibility of meteorologically informed daily screening of ground subsidence risk at a prototype level. These findings are expected to facilitate the development of a more robust ground subsidence prediction system through future integration with station-level meteorological inputs and data on subsurface infrastructure and geological conditions. Full article

As a bio-inspired paradigm fully implementable in optics, deep reservoir computing (RC) has emerged as a powerful computational framework for efficient information processing. However, conventional deep RC architectures typically demand substantial hardware overhead to establish deep nonlinear mapping. To address this, we propose a high-speed, single-physical-hardware deep photonic RC architecture based on an optically pumped spin vertical-cavity surface-emitting laser (spin-VCSEL). At the architectural level, an all-optical deep RC structure is established within a single-physical-hardware platform by feedforward-injecting the right-circularly polarized response into the left-circularly polarized mode, fully exploiting the nonlinear dynamics without another reservoir laser. Algorithmically, a compression–decompression framework is integrated to mitigate the latency overhead induced by time-division multiplexing without compromising performance. Such state reconstruction enables a tenfold reduction in the required optical-domain time-division multiplexing (TDM) processing duration. Numerical simulations demonstrate that the proposed system exhibits superior precision compared to conventional setups, yielding a normalized mean square error of 0.0039 in Santa Fe time-series prediction, a symbol error rate of 0.003 in nonlinear channel equalization, and a linear memory capacity of 19.82. Cross-correlation analysis supports reliable information transfer between the two polarization modes. Ultimately, this hardware-software co-designed paradigm offers a compact, low-latency platform for optical neuromorphic processing. Full article

Objective: To evaluate the association between daily meteorological variability and pediatric afebrile seizure presentations using ecological time-series analysis and Distributed Lag Non-linear Models (DLNMs). Methods: This retrospective ecological time-series study was conducted in a tertiary pediatric emergency department in northwestern Türkiye. Pediatric seizure-related presentations recorded between January 2019 and December 2023 were retrospectively screened for case identification and clinical validation. The primary environmental exposure analysis was restricted to eligible afebrile seizure presentations occurring during the 2023 calendar year and linked to synchronized daily meteorological data. Associations between meteorological variables and daily afebrile seizure presentation counts were assessed using multivariable Quasi-Poisson regression and Distributed Lag Non-linear Models (DLNMs). Results: Among 777 screened seizure-related presentations, 67 children contributing 157 afebrile seizure presentations during 2023 were included. Relative humidity showed the strongest independent association with pediatric afebrile seizure presentations. Each 10-percentage-point increase in relative humidity was associated with an 18.6% higher rate of pediatric afebrile seizure presentations (IRR = 1.186; 95% CI: 1.103–1.275; p < 0.001). DLNM analysis demonstrated a significant cumulative delayed association, with a 10-percentage-point increase in relative humidity associated with a 15% higher cumulative rate of afebrile seizure presentations across the 0–3-day lag period (cumulative IRR = 1.15; 95% CI: 1.08–1.22; p < 0.001). Ambient temperature and precipitation demonstrated less consistent associations across analytical approaches, whereas humidity-related effects remained robust throughout all analyses. Conclusions: Higher relative humidity was associated with increased pediatric afebrile seizure presentations and represented the only meteorological exposure demonstrating both significant contemporaneous and cumulative delayed associations. These findings suggest that meteorological variability, particularly relative humidity, may contribute to short-term and delayed fluctuations in seizure-related healthcare utilization among children. Given the single-center ecological design and the one-year environmental exposure period, these findings should be interpreted as population-level and hypothesis-generating and should not be considered a basis for individual risk prediction or humidity-based clinical counseling. Further multicenter, multi-year prospective studies incorporating individual-level and higher-resolution environmental exposure assessment are needed to confirm these findings and clarify the biological mechanisms underlying humidity-associated seizure susceptibility in pediatric populations. Full article