Search Results (327)

Forecasting cryptocurrency prices remains difficult because market dynamics are highly volatile, non-stationary, and regime-dependent. This study investigates whether combining a spiking-inspired recurrent architecture with the Grey Wolf Optimizer (GWO) can improve one-step-ahead Bitcoin forecasting within a controlled model family. We compare four configurations, LSTM, SLSTM, GWO-LSTM, and GWO-SLSTM, on 4039 daily BTC–USD closing prices from 17 September 2014 to 9 October 2025 using Min–Max normalization, strict chronological splitting, windowed regime-based robustness analysis across three distinct market regimes, and repeated-run testing. The proposed SLSTM replaces the conventional hidden-state recurrence with leaky integrate-and-fire-inspired synaptic, membrane, and adaptive-threshold dynamics, functioning as a spiking-inspired recurrent model with thresholded event gating (reset = `none’, learnable threshold). On the primary hold-out split, GWO-SLSTM achieved a test RMSE of 1840.97 and a test MAPE of 1.76%, compared with 2217.24 and 2.46% for GWO-LSTM, 3501.48 and 3.86% for SLSTM, and 4030.10 and 4.40% for LSTM. Both GWO-optimized models exhibited substantial improvements over their non-optimized counterparts, while the SLSTM baseline also outperformed the plain LSTM, indicating gains from both spiking-inspired recurrence and evolutionary hyperparameter optimization. Both optimized models exhibited near-zero bias (PBIAS 0.11% for GWO-LSTM and 0.36% for GWO-SLSTM). Within the present implementation, GWO-SLSTM also trained faster than GWO-LSTM (39.71 s vs. 137.28 s), although this runtime difference should be interpreted as setup-specific because the model families were implemented in different frameworks and stopped after different numbers of epochs. Overall, within the expanded univariate BTC–USD setting, the results support GWO-SLSTM as a strong within-family candidate for one-step-ahead forecasting under non-stationary conditions. Full article

Modern inpatient care generates irregular streams of heterogeneous clinical events, yet most predictive models require fixed feature matrices, predefined time windows, or discretization of continuous measurements. We developed CHaRT, a decoder-only autoregressive transformer designed to jointly forecast the identity of the next clinical event and, when applicable, its associated continuous value. CHaRT was trained and internally validated on structured electronic health record data from adult acute-care encounters across a 12-hospital health system in Minnesota from 2001 to 2025. The final corpus included 4,447,625 encounters from 1,301,502 patients and 701,556,877 non-padding clinical event tokens spanning vital signs, laboratory values, medications, diagnoses, microbiology, virology, imaging, fluids, and outcomes (ICU transfer or death). Encounters were split into training, validation, and test sets before vocabulary construction, normalization, and windowing. On the held-out test set, CHaRT achieved Top-1, Top-5, and Top-10 next-event accuracies of 51.61%, 87.34%, and 93.22%, respectively, with perplexity 4.50 and expected calibration error 0.0109. For numeric prediction, z-score MSE was 0.3812 for vital signs and 0.5713 for laboratory values. Seeded examples generated clinically coherent trajectories. Using model representations, a linear probe predicted deterioration (ICU transfer or in-hospital death) at a 6 h landmark with AUROC 0.95–0.97, indicating that learned representations transfer to downstream clinical risk prediction. Full article

Field hyperspectral sensing can estimate crop biomass, but model ranking may depend strongly on validation design. We evaluated winter wheat aboveground dry biomass (AGDB) estimation using 84 canopy spectra collected across two growing seasons and seven nitrogen-management treatments in Shanxi, China. Six spectral inputs were compared with CARS-UVE band selection, partial least squares regression (PLSR), and support vector regression (SVR). Under a conventional 70/30 pooled split, SG + CARS-UVE + SVR gave the highest apparent accuracy (R² = 0.8864, RMSE = 0.1174 kg m⁻², RPD = 2.9665). This advantage was not stable. Across 20 SG-based repeated splits, CARS-UVE-SVR reached a mean R² of 0.7413 with a 95% confidence interval of 0.6941–0.7885, similar to full-band PLSR (0.7448, 0.7058–0.7837), and pairwise tests showed no significant R² advantage. Cross-year transfer further favored simpler latent-variable models: SG + CARS-UVE + PLSR reached R² = 0.7577 in the 2021 → 2022 direction, whereas the pooled best SVR model dropped to R² = 0.3402. A stricter same-window cross-year analysis produced weak or negative R² values, showing that broad phenological biomass gradients supported much of the pooled accuracy. Recurrent selected regions occurred near 436–441 nm, 506–516 nm, and 711–713 nm. These findings suggest that repeated and transfer-oriented validation should be used routinely before hyperspectral biomass models are interpreted for cross-season crop monitoring. Full article

Land surface temperature (LST) is a key variable in the physics of land surface processes on both regional and global scales. This paper addresses the all-sky (clear-sky and cloudy-sky) LSTs retrieval from the data acquired by the Medium-Resolution Spectral Imager II on Fengyun 3D (FY-3D) satellite. First, an improved split-window algorithm to retrieve clear-sky LSTs is developed using numerical radiative transfer modeling experiments. Then, clear-sky LSTs are retrieved from MERSI-II/FY-3D data in January and July 2022 over an Asian area (70°E~130°E, 10°N~50°N), and cross-validated against MODIS/Aqua LST/emissivity (LST/E) Daily version 6 (MYD11C1 V6) product. Next, a hybrid method combining the eXtreme Gradient Boosting (XGBoost) model and the surface energy balance theory is developed to estimate cloudy-sky LSTs. After that, cloudy-sky LSTs are estimated from the MERSI-II data and validated with the China Meteorological Administration Land Data Assimilation System Version 2 (CLDAS V2) dataset. Against the MYD11C1 LSTs, the root mean square error (RMSE), bias and coefficient of determination (R²) of the retrieved clear-sky LSTs are 1.15 K, 0.01 ± 1.14 K, and 0.99, respectively. Against the CLDAS LSTs, the RMSE, bias and R² of the estimated hypothetical clear-sky LSTs are 4.05 K, 0.75 ± 3.98 K and 0.91, respectively, while they are 3.69 K, 0.36 ± 3.67 K, and 0.92 for the retrieved cloudy-sky LSTs, respectively, which indicates that the retrieval accuracy of cloudy-sky LSTs is improved after the cloud radiation effect correction. The all-sky LSTs retrieved in this study are accurate and consistent with the results in previous studies. Full article

Autonomous unmanned aerial vehicles (UAVs) must decide when to trust onboard perception, when to request edge support, and when to avoid acting under poor visual or communication conditions. This study develops a risk-aware edge-assisted UAV perception framework that combines calibrated visual confidence with next-window service-level agreement (SLA) feasibility. The local branch uses MobileNetV3-Small for fast onboard color recognition, while the edge branch uses ResNet-18 for stronger remote inference. Low-confidence samples are offloaded only when the SLA predictor estimates that the wireless link is feasible; otherwise, the system enters fallback, meaning that the current prediction is not treated as immediately actionable. The evaluation follows a hard cross-illumination split: indoor and fluorescent light samples are used for training and validation, and indoor night and sunlight samples are reserved for testing. Under this setting, the local model achieves 76.89% accuracy and 73.25% macro-F1, while the edge model achieves 81.26% accuracy and 77.58% macro-F1. The SLA predictor, trained on enhanced telemetry features while preserving the original target label, achieves 85.74% accuracy, 85.57% macro-F1, 0.9420 ROC-AUC, and 0.9585 PR-AUC on temporally held-out records. The joint policy achieves 93.23% coverage and 79.90% success over active decisions, using local inference for 82.76% of the samples, edge offloading for 10.47%, and fallback for 6.77%. These results indicate that the framework is best understood as a tunable risk management layer for UAV perception rather than a pure accuracy maximization classifier. It avoids blind offloading and reduces forced decisions when both visual confidence and communication feasibility are weak. Full article

Short-term wind power forecasting is a key enabling technology for wind farm operation optimization, power grid dispatch, and electricity market decision-making. However, existing studies often lack unified standards in data partitioning, input feature construction, and hyperparameter tuning, making fair and reproducible comparisons across models difficult to achieve. To address this issue, this study focuses on day-ahead wind power forecasting for a single wind farm and establishes a benchmarking framework with strict chronological splitting, a shared feature information set, and a consistent hyperparameter tuning budget. Within this framework, six representative models, namely Ridge, XGBoost, LightGBM, DLinear, Transformer, and PatchTST, are systematically evaluated. A two-level evaluation protocol combining a fixed hold-out split and expanding-window rolling validation is adopted to compare model performance from multiple perspectives, including overall accuracy, sensitivity to hyperparameter tuning, robustness across rolling windows, and performance under typical operating scenarios. The results show that model rankings are not fully consistent between the hold-out evaluation and the rolling-validation setting. Under the fixed hold-out split, LightGBM achieved the lowest NRMSE of 10.2326%, while Transformer obtained the lowest NMAE of 6.9944%. In contrast, under the 8-fold expanding-window rolling validation, Transformer achieved the lowest average NRMSE of 8.1684%, followed by LightGBM with 8.7344%. These results indicate that the best performance on a single test split does not necessarily imply the strongest robustness across multiple time windows. In addition, strong tree-based models remain highly competitive in this single-wind-farm forecasting task, whereas more complex deep temporal models do not always deliver stable advantages. Meanwhile, the performance gains brought by hyperparameter optimization vary substantially across models, suggesting that conclusions drawn from default-parameter comparisons are of limited reliability. These findings demonstrate that systematic benchmarking under strict temporal constraints and fair tuning conditions is essential for clarifying the comparative performance, robustness, and engineering applicability of different model families. The study can therefore provide practical guidance for model selection and deployment in short-term wind power forecasting for single wind farms. Full article

This study presents a proof-of-concept investigation into the use of proximal policy optimization (PPO), a deep reinforcement learning (DRL) algorithm, for estimating the remaining useful life (RUL) of rolling element bearings. Although DRL has shown growing promise in prognostics, existing applications have predominantly relied on off-policy deterministic actor–critic methods such as deep deterministic policy gradient (DDPG) and twin delayed DDPG (TD3); the suitability of on-policy clipped-objective methods such as PPO for this task remains comparatively unexplored. To address this gap, statistical time-domain features are extracted from raw vibration signals and used as input to train a PPO agent with an actor–critic architecture, in which the actor network predicts RUL values and the critic network evaluates prediction quality through state-value estimation. A preprocessing pipeline comprising feature extraction, normalization, and sliding-window segmentation is developed, and the PPO framework incorporates generalized advantage estimation (GAE), a custom-designed reward function, and a policy-clipping mechanism to support stable training. The method is evaluated on a representative bearing (Bearing 2_1) from the XJTU-SY run-to-failure dataset using a chronological train/test split, and benchmarked against long short-term memory (LSTM) networks, multilayer perceptrons (MLPs), and a naive linear regression baseline. Performance is assessed using root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), and a domain-specific asymmetric scoring function that penalizes late predictions more heavily than early ones. Experimental results show that the PPO-based model produces more stable and operationally favourable RUL estimates than the supervised baselines on the unseen late-degradation segment, particularly in the critical end-of-life region. The findings support PPO as a viable on-policy DRL formulation for bearing RUL prediction and motivate further validation across multiple bearings and operating conditions, identified here as essential future work. Full article

Distributed seismic sensor networks integrated into the Internet of Things (IoT) infrastructure enable continuous condition monitoring of large-scale engineering structures. During long-term operation, however, measurement channels are subject to sensitivity drift, increased noise, and pulse artifacts that statistically mimic real vibration events. Related deep-learning techniques for noisy and ill-posed inverse problems have demonstrated the value of combining principled physical priors with deep models. Although the application domain differs, the underlying methodological insight—that constrained, physics-aware feature mappings can stabilize learning under noisy and partially observed conditions—directly motivates the use of a parameterized quantum circuit as a nonlinear feature transformer in the present work, where Hilbert space mapping serves as an analogous structural prior for the latent representation. Three principal fault modes are considered in this work, corresponding to the dominant degradation mechanisms observed in long-term seismic instrumentation: sensor drift, increased noise, and sensor failure. Each fault mode produces a distinct signature in the windowed feature space; the proposed model is trained to discriminate between them based on the latent CNN-LSTM-VQC representation. We propose a hybrid quantum-inspired deep-learning model (QC-DL) for the detection and diagnosis of channel-degradation anomalies. The architecture combines a 1D-CNN+LSTM feature extractor with a parameterized variational quantum circuit (VQC) used as a nonlinear feature transformer. All quantum experiments were performed on the QPanda3 CPUQVM simulator. The data were split chronologically prior to windowing to avoid information leakage. On real-world labeled accelerometric data with four operating modes (normal/drift/high-noise/failure), the QC-DL model achieved a macro-averaged F1 score of approximately 0.69 and per-class AUC values in the range 0.88–0.99. The mean early-detection latency was 1.6 s versus 2.1 s for the CNN-LSTM baseline (~24% reduction). An ablation study against a parameter-matched classical MLP showed that the gain is modest and not solely attributable to additional nonlinearity. The reported p-values (p = 0.70, p = 0.29) do not establish statistical significance. The results support the feasibility of hybrid quantum-inspired deep learning for sensor-channel verification, while highlighting the need for evaluation on real NISQ hardware. This paper proposes a hybrid quantum-inspired approach for detecting and diagnosing such anomalies in the time series of distributed seismic networks. The architecture combines a classical temporal feature extraction module based on one-dimensional convolutional layers and a recurrent long short-term memory (LSTM) network, which generates a latent window representation of the signal, with a parameterized variational quantum circuit used as a nonlinear feature processor in a hybrid computational circuit. Experimental validation was performed on real-world labeled data with multiple sensor degradation modes. The evaluation was organized in a scoring framework aligned with autonomous operation through window ranking and threshold alarm generation. In the experiments, the proposed model provided a macro-averaged F1 score of approximately 0.69 and area under the receiver operating characteristic (AUC) curve values in the range of 0.88–0.99 across classes, outperforming baseline deep models. The average early detection latency was 1.6 s versus 2.1 s for the baseline recurrent model (a 24% reduction). An ablative comparison with a control model based on a classical multilayer perceptron of comparable dimension confirmed that the improvement is not limited to the addition of additional nonlinearity. The obtained results indicate the potential of quantum-supported deep learning for improving the reliability of long-term vibration monitoring and verifying the correctness of sensor channels in distributed seismic networks. Full article

Industrial control systems (ICS) are increasingly targeted by cyberattacks that manipulate physical processes while evading data-driven detectors trained on raw time-series data. This paper extracts 34–41 control-theoretic features, including tracking error, valve mismatch, sensor liveness, and their temporal derivatives, from Proportional–Integral–Derivative (PID) control loops and evaluates them using an Isolation Forest combined with a maximum z-score. On HAI 21.03, Stage 1 achieves a PA-F1 score of 0.8945, detecting 48 out of 50 attacks. On HAI 23.05, Stage 1 attains a PA-F1 score of 0.9210, surpassing seven deep-learning baselines by at least 23 PA-F1 points; the closest baseline, a learned Graph Neural Network (GNN), achieves 0.6890. Re-implementations of ConvBiLSTM-AE (PA-F1 = 0.6689) and TranAD (PA-F1 = 0.6838) on the same evaluation split confirm this performance gap. A controlled USAD experiment, with PA-F1 = 0.7343 for physics features versus 0.6687 for raw Supervisory Control and Data Acquisition (SCADA), demonstrates that the extracted features provide the detection signal independently of the model architecture. Adding a bidirectional Gated Recurrent Unit (GRU) refinement stage improves PA-F1 by 8.1 percentage points on HAI 21.03, but the same stage reduces it by 6.8 percentage points on HAI 23.05, where attacks manifest as brief perturbations; four alternative Stage 2 designs reproduce this degradation. We therefore characterize temporal refinement as beneficial only for sustained-deviation attacks and identify Stage 1 as the primary deployable detector. This study is the first to apply physics-informed features, report both PA-F1 and eTaPR on HAI 23.05, and perform per-window error diagnosis on this dataset. Results show that 10 of 15 detected windows are covered by fewer than 10% of their timesteps, revealing a structural tension between PA-F1 and eTaPR. Full article

Pipelines remain the primary mode of oil and gas transportation but are vulnerable to leaks that pose environmental and safety risks, particularly in two-phase flow systems. Conventional detection methods often struggle under transient multiphase conditions, while many data-driven studies rely on static evaluation metrics that do not reflect continuous monitoring requirements. This study develops a machine learning framework for leak detection using OLGA-simulated datasets from a previously published study, comprising approximately 180,000 labelled samples across nine leak scenarios and one no-leak case. Pressure, temperature, and mass-flow variables were enhanced through feature engineering to capture nonlinear leak behaviour. Random forest and extreme gradient boosting (XGBoost) classifiers were trained using an 80/20 stratified split with synthetic minority oversampling technique (SMOTE)-based balancing applied only to training data. XGBoost achieved 99.2% accuracy and reduced false positives by 53% relative to random forest while maintaining near-zero false negatives. A sliding-window suspicion framework extended static classification into time-dependent detection, producing delays of between 9.81 s and 82.04 s with zero false alarms in the no-leak scenario. Physical validation using pressure, flow, and fast Fourier transform (FFT) analysis confirmed that detections correspond to genuine hydraulic disturbances, demonstrating the reliability and physical credibility of the proposed framework. Full article

Trustworthy deployment of artificial intelligence in safety-critical systems requires accurate diagnosis of anticipated scenarios and reliable rejection of out-of-distribution (OOD) inputs that fall outside the modeled operational scope. Existing data-driven diagnostic models typically assume that test inputs are drawn from the training distribution or rely on heuristically tuned thresholds that lack enforceable safety guarantees. This article presents SCOPE (Safety-Calibrated Out-of-distribution Prediction via Contrastive Embeddings), a framework integrating supervised contrastive learning with split-conformal prediction to provide statistically grounded OOD rejection with finite-sample false-alarm control. SCOPE employs a causal residual convolutional encoder to map multivariate sensor streams into a hyperspherical embedding space with a compact, class-specific structure. A k-nearest-neighbor density nonconformity score, computed in the encoder embedding space, flags transients that occupy low-density regions relative to known accident manifolds; an ablation shows that this density score outperforms prototype distance, entropy, and conservative maximum fusion as well as a panel of standard OOD baselines (MSP, ODIN, energy, Mahalanobis, OpenMax, MC-dropout, and a reconstruction autoencoder). To support temporally evolving trajectories, SCOPE aggregates window-level scores under a monotone decision policy and performs trajectory-level conformal calibration, yielding distribution-free guarantees that bound the probability of falsely rejecting a known accident run. SCOPE is evaluated on the Nuclear Power Plant Accident Data (NPPAD) benchmark using high-openness splits that withhold entire accident families as unknowns, and all metrics are reported as mean ± standard deviation across multiple random seeds. Results demonstrate strong diagnostic accuracy on accepted trajectories, conservative false-alarm rates satisfying user-specified safety constraints across multiple operating points, and timely rejection of unseen accident mechanisms, making SCOPE suitable for deployment in safety-critical monitoring applications. Full article

Identifying insider threats in modern enterprise environments presents a unique cybersecurity challenge. Although malicious activity may often appear to be legitimate user activity, it is difficult to recognize the distinction. This study presents an innovative approach to insider threat detection by analyzing enterprise activity logs for session-level behavioural risk monitoring with behavioural biometrics. Behavioural patterns are modelled as temporal sequences across consecutive monitoring windows to capture both short-term behavioural intensity and long-term behavioural drift. The proposed system utilizes a hybrid deep learning architecture that includes a Long Short-Term Memory (LSTM) network and an autoencoder model to model temporal dependence of a user’s behaviour and to identify anomalies through reconstruction error analysis. The LSTM network captures user’s sequential activity and autoencoder determines variance from the user’s typical behavioural profile. The outputs of both models are aggregated using a unified behavioural risk scoring mechanism for session-level risk monitoring and ongoing insider threat assessment. The experimental results from Insider Threat Dataset for Corporate Environments demonstrate that proposed approach is effective in classifying normal versus malicious behaviours of users. The proposed framework achieves an accuracy of 97.65%, a precision of 96.35%, a recall of 99.05%, an F1-score of 97.68%, and a ROC-AUC of 99.20% on a near-balanced benchmark split. Under realistic class imbalance conditions, the framework achieves a PR-AUC of 0.842 and MCC of 0.781, representing the more operationally conservative performance estimate. These findings confirm that the proposed framework constitutes a viable solution for integrating behavioural modelling and anomaly detection within continuous enterprise authentication systems. Full article

Temporary container buildings are widely used because of their rapid construction, flexible deployment, and suitability for construction-site accommodation, emergency facilities, event housing, and other short-term scenarios. However, their energy-saving design still lacks specialized standards. Key parameters such as insulation thickness, window thermal performance, airtightness, and split-air-conditioner efficiency are often selected empirically, which makes it difficult to balance initial investment and operating cost over the actual service life. To address these issues, this study proposes a service-life cost-based coordinated optimization framework. The framework couples DeST hourly load simulation, a TRNSYS-derived dynamic energy-efficiency-ratio (EER) model for split-type air conditioners, an economic model including initial investment and electricity operating cost, and an SLSQP-based optimizer. Field measurements from a three-story container dormitory in Haidian District, Beijing, collected in August and December 2023, are used to validate the HVAC electricity-consumption model through cumulative electricity-consumption errors and CV(RMSE). Using a south-facing single container building in Beijing as the base case, optimization is conducted for design service lives of 1–10 years and further compared under different electricity-pricing models and climate regions. The results show that, within the allowable parameter ranges, the proposed method can reduce service-life cost by up to approximately 32%. In the Beijing 2-year case, the optimized scheme reduces service-life cost by 39.9% compared with the permanent-building-code benchmark and by 11.4% compared with a market sample. The results demonstrate that coordinated envelope–HVAC optimization can avoid redundant initial investment and provide scenario-adaptable design support for temporary container buildings. Full article

Induction motors are critical in modern industry, powering over 70% of industrial processes. Reliable operation is essential to minimize downtime and ensure production continuity. This paper proposes an integrated multimodal methodology for fault diagnosis and prognosis in induction motors, based on an extended Pearson and Gain feature fusion framework. The approach preprocesses vibration, current, voltage, torque, and speed signals through denoising, normalization, synchronization, and sliding-window segmentation. Over 200 features per window are extracted across time, frequency, envelope, wavelet, harmonic, slip-based, and MCSA domains. A key innovation is correlation-driven multimodal fusion, combining Pearson correlation, spectral coherence, cross-spectral energy, and mutual information to produce Gain-enhanced features with improved discriminative capability. Fault diagnosis is performed using RF, SVM, XGBoost, and MLP models, with time-aware data splitting to avoid temporal leakage. Prognosis employs a continuous Degradation Index (DI) modeled via Gaussian Process Regression for uncertainty-aware prediction, with failure probability and Remaining Useful Life (RUL) estimated from DI thresholds. Experimental results demonstrate that the proposed methodology achieves diagnostic accuracy above 97%, enhances feature relevance, and provides stable long-term prognostic performance, offering a robust framework for predictive maintenance of induction motors. Full article

Urban traffic accidents emerge from complex interactions among traffic instability, roadway structure, environmental disturbance, and temporal dynamics, yet many existing prediction approaches still treat accident risk as a discrete classification problem over isolated observations. This study proposes a Continuous Hazard Field Intelligence Framework for Traffic Accident Emergence and Urban Safety Prediction, which models hidden urban danger as a topology-aware spatio-temporal hazard field that evolves continuously across connected transportation infrastructure. The framework integrates heterogeneous urban traffic observations, including incident records, crash data, roadway attributes, temporal cues, and contextual risk factors, into a unified hazard-aware learning pipeline. A dedicated preprocessing strategy combines topology-constrained spatial alignment, temporal hazard window embedding, risk-diffusion feature lifting, hazard-sensitive normalization, and continuous hazard surface initialization to convert fragmented event-centered observations into a smooth and learning-ready hazard representation. A structured deep learning architecture is then developed to perform spatial hazard encoding, temporal hazard evolution, continuous hazard reconstruction, and localized accident emergence prediction. Experimental evaluation was conducted on two large-scale real-world traffic safety datasets, namely the XTraffic Incident Dataset (2022–2024) with 1,441,904 records and the Motor Vehicle Collisions–Crashes Dataset with 2,026,647 records. All model configurations were evaluated under the same experimental setting, using the same dataset-specific preprocessing protocol, a 70/30 train–test split, and identical evaluation metrics. The final CHFI configuration achieves 99.12% accuracy, 98.94% precision, 98.76% recall, 98.85% F1-score, and 0.998 AUC on Dataset 1, and 98.63% accuracy, 98.41% precision, 98.16% recall, 98.28% F1-score, and 0.997 AUC on Dataset 2. Compared with the initial non-hazard-aware baseline configuration evaluated under the same data split and evaluation protocol, the final CHFI model improves the F1-score by 7.91 percentage points on Dataset 1 and 8.26 percentage points on Dataset 2. These results indicate that the proposed hazard-field formulation can improve accident-emergence prediction within the controlled experimental setting, while the reported gains should be interpreted relative to the specified baseline and evaluation design. Full article