Search Results (3,186)

Adversarial attacks in Natural Language Processing (NLP) present a critical challenge, particularly in sentiment analysis, where subtle input modifications can significantly alter model predictions. In search of more robust defenses against adversarial attacks on sentimental analysis, this research work introduces two novel defense mechanisms: the Lexicon-Based Random Substitute Model (LRSM) and the Word-Variant Voting Model (WVVM). LRSM employs randomized substitutions from a dataset-specific lexicon to generate diverse input variations, disrupting adversarial strategies by introducing unpredictability. Unlike traditional defenses requiring synonym dictionaries or precomputed semantic relationships, LRSM directly substitutes words with random lexicon alternatives, reducing overhead while maintaining robustness. Notably, LRSM not only neutralizes adversarial perturbations but occasionally surpasses the original accuracy by correcting inherent model misclassifications. Building on LRSM, WVVM integrates LRSM, Frequency-Guided Word Substitution (FGWS), and Synonym Random Substitution and Voting (RS&V) in an ensemble framework that adaptively combines their outputs. Logistic Regression (LR) emerged as the optimal ensemble configuration, leveraging its regularization parameters to balance the contributions of individual defenses. WVVM consistently outperformed standalone defenses, demonstrating superior restored accuracy and F1 scores across adversarial scenarios. The proposed defenses were evaluated on two well-known sentiment analysis benchmarks: the IMDB Sentiment Dataset and the Yelp Polarity Dataset. The IMDB dataset, comprising 50,000 labeled movie reviews, and the Yelp Polarity dataset, containing labeled business reviews, provided diverse linguistic challenges for assessing adversarial robustness. Both datasets were tested using 4000 adversarial examples generated by established attacks, including Probability Weighted Word Saliency, TextFooler, and BERT-based Adversarial Examples. WVVM and LRSM demonstrated superior performance in restoring accuracy and F1 scores across both datasets, with WVVM excelling through its ensemble learning framework. LRSM improved restored accuracy from 75.66% to 83.7% when compared to the second-best individual model, RS&V, while the Support Vector Classifier WVVM variation further improved restored accuracy to 93.17%. Logistic Regression WVVM achieved an F1 score of 86.26% compared to 76.80% for RS&V. These findings establish LRSM and WVVM as robust frameworks for defending against adversarial text attacks in sentiment analysis. Full article

Like in many countries, the transfer of increased land value created by public actions without landowner contributions back to the public is under debate in Türkiye. Although various Land Value Capture (LVC) mechanisms are employed worldwide to finance infrastructure investments, no comprehensive system has been established in Türkiye for this purpose. In this study, an improved LVC model that integrates land value and development rights is proposed. This model, termed Hybrid Land Readjustment (hLR), is designed to ensure that land value increases triggered by public investments are returned to the public. To this end, existing Turkish value capture instruments with potential are examined. Under the proposed hLR framework, equal basic development rights are granted to cadastral parcels, parcel and building-block value maps are utilized, basic rights are adjusted according to land-value changes, and a portion of additional development rights is transferred to the public. A practical application scenario is provided to illustrate the model’s operation. The system is configured for seamless integration into Türkiye’s existing legal and planning framework, offering a sustainable mechanism for financing infrastructure and implementing zoning plans. Full article

The Mw 7.8 Gorkha Earthquake of 25 April 2015 triggered over 25,000 landslides across central Nepal, with 4775 events concentrated in Gorkha District alone. Despite substantial advances in landslide susceptibility mapping, existing studies often overlook the compound role of post-seismic rainfall and lack robust spatial validation. To address this gap, we validated an ensemble machine learning framework for co-seismic landslide susceptibility modeling by integrating seismic, geomorphological, hydrological, and anthropogenic variables, including cumulative post-seismic rainfall. Using a balanced dataset of 4775 landslide and non-landslide instances, we evaluated the performance of Logistic Regression (LR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) models through spatial cross-validation, SHapley Additive exPlanations (SHAP) explainability, and ablation analysis. The RF model outperformed all others, achieving an accuracy of 87.9% and a Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) value of 0.94, while XGBoost closely followed (AUC = 0.93). Ensemble models collectively classified over 95% of observed landslides into High and Very High susceptibility zones, demonstrating strong spatial reliability. SHAP analysis identified elevation, proximity to fault, peak ground acceleration (PGA), slope, and rainfall as dominant predictors. Notably, the inclusion of post-seismic rainfall substantially improved recall and F1 scores in ablation experiments. Spatial cross-validation revealed the superior generalizability of ensemble models under heterogeneous terrain conditions. The findings underscore the value of integrating post-seismic hydrometeorological factors and spatial validation into susceptibility assessments. We recommend adopting ensemble models, particularly RF, for operational hazard mapping in earthquake-prone mountainous regions. Future research should explore the integration of dynamic rainfall thresholds and physics-informed frameworks to enhance early warning systems and climate resilience. Full article

Previous research has primarily focused on assessing seasonal mean or annual extreme climate events, whereas intraseasonal variability in extreme climate has received comparatively little attention, despite its importance for understanding short-term climate dynamics and associated risks. This study evaluates the performance of nine climate models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) in reproducing summer maximum temperature (Tmax) variability across China during 1979–2014, with the variability defined as the standard deviation of daily Tmax anomalies for each summer. Results show that most CMIP6 models fail to reproduce the observed north–south gradient of Tmax variability with significant regional biases and limited agreement on temporal trends. The multi-model ensemble (MME) outperforms most individual models in terms of root-mean-square error and spatial correlation, but it still under-represents the observed temporal trends, especially over southeastern and central China. Taylor diagram analysis reveals that EC-Earth3, GISS-E2-1-G, IPSL-CM6A-LR, and the MME perform relatively well in capturing the spatial characteristics of Tmax variability, whereas MIROC6 shows the poorest performance. These findings highlight the persistent limitations in simulating intraseasonal Tmax variability and underscore the need for improved model representations of regional climate dynamics over China. Full article

Background: Overcrowded emergency departments (EDs) create significant challenges for patient management and hospital efficiency. In response, Amiens Picardy University Hospital (APUH) developed the “Prediction of the Patient Pathway in the Emergency Department” (3P-U) model to enhance patient flow management. Objectives: To develop and clinically validate an explainable artificial intelligence (XAI) model for hospital admission predictions, using structured triage data, and demonstrate its real-world applicability in the ED setting. Methods: Our retrospective, single-center study involved 351,019 patients consulting in APUH’s EDs between 2015 and 2018. Various models (including a cross-validation artificial neural network (ANN), a k-nearest neighbors (KNN) model, a logistic regression (LR) model, and a random forest (RF) model) were trained and assessed for performance with regard to the area under the receiver operating characteristic curve (AUROC). The best model was validated internally with a test set, and the F1 score was used to determine the best threshold for recall, precision, and accuracy. XAI techniques, such as Shapley additive explanations (SHAP) and partial dependence plots (PDP) were employed, and the clinical explanations were evaluated by emergency physicians. Results: The ANN gave the best performance during the training stage, with an AUROC of 83.1% (SD: 0.2%) for the test set; it surpassed the RF (AUROC: 71.6%, SD: 0.1%), KNN (AUROC: 67.2%, SD: 0.2%), and LR (AUROC: 71.5%, SD: 0.2%) models. In an internal validation, the ANN’s AUROC was 83.2%. The best F1 score (0.67) determined that 0.35 was the optimal threshold; the corresponding recall, precision, and accuracy were 75.7%, 59.7%, and 75.3%, respectively. The SHAP and PDP XAI techniques (as assessed by emergency physicians) highlighted patient age, heart rate, and presentation with multiple injuries as the features that most specifically influenced the admission from the ED to a hospital ward. These insights are being used in bed allocation and patient prioritization, directly improving ED operations. Conclusions: The 3P-U model demonstrates practical utility by reducing ED crowding and enhancing decision-making processes at APUH. Its transparency and physician validation foster trust, facilitating its adoption in clinical practice and offering a replicable framework for other hospitals to optimize patient flow. Full article

Secure autonomous secret key distillation (SKD) systems traditionally depend on external common randomness (CR) sources, which often suffer from instability and limited reliability over long-term operation. In this work, we propose a novel SKD architecture that synthesizes CR by combining a keystream of a shared-key keystream generator $KSG\left(K_G\right)$ with locally generated binary Bernoulli noise. This construction emulates the statistical properties of the classical Maurer satellite scenario while enabling deterministic control over key parameters such as bit error rate, entropy, and leakage rate (LR). We derive a closed-form lower bound on the equivocation of the shared-secret key $K_G$ from the viewpoint of an adversary with access to public reconciliation data. This allows us to define an admissible operational region in which the system guarantees long-term secrecy through periodic key refreshes, without relying on advantage distillation. We integrate the Winnow protocol as the information reconciliation mechanism, optimized for short block lengths ($N=8$), and analyze its performance in terms of efficiency, LR, and final key disagreement rate (KDR). The proposed system operates in two modes: ideal secrecy, achieving secret key rates up to 22% under stringent constraints (KDR < 10⁻⁵, LR < 10⁻¹⁰), and perfect secrecy mode, which approximately halves the key rate. Notably, these security guarantees are achieved autonomously, without reliance on advantage distillation or external CR sources. Theoretical findings are further supported by experimental verification demonstrating the practical viability of the proposed system under realistic conditions. This study introduces, for the first time, an autonomous CR-based SKD system with provable security performance independent of communication channels or external randomness, thus enhancing the practical viability of secure key distribution schemes. Full article

In mining areas with high groundwater levels, intensive coal mining has led to the accumulation of substantial surface water and significant alterations in regional landscape patterns. Reconstructing the ecological security pattern (ESP) has emerged as a critical focus for ecological restoration in coal mining subsidence areas with high groundwater levels. This study employed the patch-generating land use simulation (PLUS) model to predict the landscape evolution trend of the study area in 2032 under three scenarios, combining environmental characteristics and disturbance features of coal mining subsidence areas with high groundwater levels. In order to determine the differences in ecological network changes within the study area under various development scenarios, morphological spatial pattern analysis (MSPA) and landscape connectivity analysis were employed to identify ecological source areas and establish ecological corridors using circuit theory. Based on the simulation results of the optimal development scenario, potential ecological pinch points and ecological barrier points were further identified. The findings indicate that: (1) land use changes predominantly occur in urban fringe areas and coal mining subsidence areas. In the land reclamation (LR) scenario, the reduction in cultivated land area is minimal, whereas in the economic development (ED) scenario, construction land exhibits a marked increasing trend. Under the natural development (ND) scenario, forest land and water expand most significantly, thereby maximizing ecological space. (2) Under the ND scenario, the number and distribution of ecological source areas and ecological corridors reach their peak, leading to an enhanced ecological network structure that positively contributes to corridor improvement. (3) By comparing the ESP in the ND scenario in 2032 with that in 2022, the number and area of ecological barrier points increase substantially while the number and area of ecological pinch points decrease. These areas should be prioritized for ecological protection and restoration. Based on the scenario simulation results, this study proposes a planning objective for a “one axis, four belts, and four zones” ESP, along with corresponding strategies for ecological protection and restoration. This research provides a crucial foundation for decision-making in enhancing territorial space planning in coal mining subsidence areas with high groundwater levels. Full article

Temperature monitoring in extreme environments, such as coal-fired power plants, was addressed by designing and testing wireless patch antennas for use in machine learning-aided temperature estimation. The sensors were designed to monitor the temperature and health of boiler systems. Wireless interrogation of the sensor was performed using a Vector Network Analyzer (VNA) and a pair of interrogation antennas to capture resonance behavior under varying thermal and spatial conditions with sensitivities ranging from 0.052 to 0.20 ${\displaystyle \frac{MHz}{°\mathrm{C}}}$. Sensor calibration was conducted using a Long Short-Term Memory (LSTM) model, which leveraged temporal patterns to account for hysteresis effects. The calibration method demonstrated improved performance when combined with an LSTM model, achieving up to a 76% improvement in temperature estimation error when compared with Linear Regression (LR). The experiments highlighted an innovative solution for patch antenna-based non-contact temperature measurement, which addresses limitations with conventional methods such as RFID-based systems, infrared, and thermocouples. Full article

Conventional ground reaction force (GRF) and load rate (LR) analyses may overlook temporal and waveform characteristics that reflect injury status and acuity. This study used an alternative GRF processing methodology to characterize GRF waveforms among runners with symptomatic medial tibial stress fractures (MTSS) and those recovering from tibial stress fractures (TSF; both unilateral [UL] and bilateral [BL]). This cross-sectional analysis of runners (n = 66) included four groups: symptomatic MTSS, recovering from UL or BL TSF, or uninjured case-matched controls. Participants ran at self-selected speed on an instrumented treadmill. Kinematics were collected with a 3D optical motion analysis system. Double-Gaussian models described the biphasic loading pattern of running gait (initial impact, active phases). Gaussian parameters described relative differences in the GRF waveform by injury condition. LR was calculated using the central difference numerical derivative of the raw normalized net force data. During the impact phase (0–20% of stance), controls and BL TSF produced higher GRF amplitudes than UL TSF and MTSS (p < 0.05). BL TSF and controls had greater maximal positive LR and minimum LR than UL TSF and MTSS. Peak medial GRF was 18–43% higher in the BL TSF group than in MTSS and UL TSF (p < 0.05). Correlations existed between tibial pain severity and early stance net GRF (r = 0.512; p = 0.016) and between pain severity and the duration since diagnosis for LR values during the impact phase (r values = 0.389–0.522; all p < 0.05). Collectively, these data suggest that this waveform modeling approach can differentiate injury status and pain acuity in runners. Early stance GRF and LR may offer novel insight into the management of running-related injuries. Full article

We present a neural-network-based approach to solve the image-to-image translation problem in microscale droplets squeeze flow. A residual convolutional neural network is proposed to address the inverse problem: reconstructing a low-resolution (LR) droplet pattern image from a high-resolution (HR) liquid film thickness imprint. This enables the prediction of initial droplet configurations that evolve into target HR imprints after a specified spreading time. The developed neural network architecture aims at learning to tune the refinement level of its residual convolutional blocks by using function approximators that are trained to map a given film thickness to an appropriate refinement level indicator. We use multiple stacks of convolutional layers, the output of which is translated according to the refinement level indicators provided by the directly connected function approximators. Together with a non-linear activation function, the translation mechanism enables the HR imprint image to be refined sequentially in multiple steps until the target LR droplet pattern image is revealed. We believe that this work holds value for the semiconductor manufacturing and packaging industry. Specifically, it enables desired layouts to be imprinted on a surface by squeezing strategically placed droplets with a blank surface, eliminating the need for customized templates and reducing manufacturing costs. Additionally, this approach has potential applications in data compression and encryption. Full article

Objective: Shoulder dystocia (ShD) is a rare but serious obstetric emergency associated with significant neonatal morbidity. This study aimed to evaluate the predictive performance of machine learning (ML) models based on fetal biometric ratios and clinical characteristics for the identification of ShD in pregnancies without clinical suspicion of macrosomia. Methods: We conducted a retrospective case-control study including 284 women (84 ShD cases and 200 controls) who underwent spontaneous vaginal delivery between 37 and 42 weeks of gestation. All participants had an estimated fetal weight (EFW) below the 90th percentile according to Hadlock reference curves. Univariate and multivariate logistic regression analyses were performed on maternal and neonatal parameters, and statistically significant variables (p < 0.05) were used to construct adjusted odds ratio (aOR) models. Supervised ML models—Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGB)—were trained and tested to assess predictive accuracy. Performance metrics included AUC-ROC, sensitivity, specificity, accuracy, and F1-score. Results: The BPD/AC ratio and AC/FL ratio markedly enhanced the prediction of ShD. When added to other features in RF models, the BPD/AC ratio got an AUC of 0.884 (95% CI: 0.802–0.957), a sensitivity of 68%, and a specificity of 83%. On the other hand, the AC/FL ratio, along with other factors, led to an AUC of 0.896 (95% CI: 0.805–0.972), 68% sensitivity, and 90% specificity. Conclusions: In pregnancies without clinical suspicion of macrosomia, ML models integrating fetal biometric ratios with maternal and labor-related factors significantly improved the prediction of ShD. These models may support clinical decision-making in low-risk deliveries where ShD is often unexpected. Full article

Temperature prediction plays a crucial role across various sectors, including agriculture and climate research. Understanding weather patterns, seasonal shifts, and climate dynamics heavily relies on accurate temperature forecasts. This paper presents an innovative hybrid method, EEMD-LR, that combines ensemble empirical mode decomposition (EEMD) with linear regression (LR) for temperature prediction. EEMD is used to decompose temperature signals into stable sub-signals, enhancing their predictability. LR is then applied to forecast each sub-signal, and the resulting predictions are integrated to obtain the final temperature forecast. The proposed EEMD-LR model achieved RMSE, MAE, and R² values of 0.000027, 0.000021, and 1.000000, respectively, on the sine simulation time-series data used in this study. For actual temperature time-series data, the model achieved RMSE, MAE, and R² values of 0.713150, 0.512700, and 0.994749, respectively. The experimental results on these two datasets indicate that the EEMD-LR model demonstrates superior predictive performance compared to alternative methods. Full article

Traditional bathymetry measures require a large number of human hours, and many bathymetry records are obsolete or missing. Automated measures of bathymetry would reduce costs and increase accessibility for research and applications. In this paper, we optimized a recent machine learning model, named CatBoostOpt, to estimate bathymetry based on high-resolution WorldView-2 (WV-2) multi-spectral optical satellite images. CatBoostOpt was demonstrated across the Florida Big Bend coastline, where the model learned correlations between in situ sound Navigation and Ranging (Sonar) bathymetry measurements and the corresponding multi-spectral reflectance values in WV-2 images to map bathymetry. We evaluated three different feature transformations as inputs for bathymetry estimation, including raw reflectance, log-linear, and log-ratio transforms of the raw reflectance value in WV-2 images. In addition, we investigated the contribution of each spectral band and found that utilizing all eight spectral bands in WV-2 images offers the best solution for handling complex water quality conditions. We compared CatBoostOpt with linear regression (LR), support vector machine (SVM), random forest (RF), AdaBoost, gradient boosting, and deep convolutional neural network (DCNN). CatBoostOpt with log-ratio transformed reflectance achieved the best performance with an average root mean square error (RMSE) of 0.34 and coefficient of determination (${\mathrm{R}}^2$) of 0.87. Full article

This study describes a quite special and interesting atmospheric event characterized by the simultaneous presence of fresh and aged smoke layers. These peculiar conditions occurred on 16 July 2024 at the CNR-IMAA atmospheric observatory (CIAO) in Potenza (Italy), and represent an ideal case for the evaluation of the impact of aging and transport mechanisms on both the optical and microphysical properties of biomass burning aerosol. The fresh smoke was originated by a local wildfire about 2 km from the measurement site and observed about one hour after its ignition. The other smoke layer was due to a wide wildfire occurring in Canada that, according to backward trajectory analysis, traveled for about 5–6 days before reaching the observatory. Synergetic use of lidar, ceilometer, radar, and microwave radiometer measurements revealed that particles from the local wildfire, located at about 3 km a.s.l., acted as condensation nuclei for cloud formation as a result of high humidity concentrations at this altitude range. Optical characterization of the fresh smoke layer based on Raman lidar measurements provided lidar ratio (LR) values of 46 ± 4 sr and 34 ± 3 sr, at 355 and 532 nm, respectively. The particle linear depolarization ratio (PLDR) at 532 nm was 0.067 ± 0.002, while backscatter-related Ångström exponent (AEβ) values were 1.21 ± 0.03, 1.23 ± 0.03, and 1.22 ± 0.04 in the spectral ranges of 355–532 nm, 355–1064 nm and 532–1064 nm, respectively. Microphysical inversion caused by these intensive optical parameters indicates a low contribution of black carbon (BC) and, despite their small size, particles remained outside the ultrafine range. Moreover, a combined use of CIAO remote sensing and in situ instrumentation shows that the particle properties are affected by humidity variations, thus suggesting a marked particle hygroscopic behavior. In contrast, the smoke plume from the Canadian wildfire traveled at altitudes between 6 and 8 km a.s.l., remaining unaffected by local humidity. Absorption in this case was higher, and, as observed in other aged wildfires, the LR at 532 nm was larger than that at 355 nm. Specifically, the LR at 355 nm was 55 ± 2 sr, while at 532 nm it was 82 ± 3 sr. The AEβ values were 1.77 ± 0.13 and 1.41 ± 0.07 at 355–532 nm and 532–1064 nm, respectively and the PLDR at 532 nm was 0.040 ± 0.003. Microphysical analysis suggests the presence of larger, yet much more absorbent particles. This analysis indicates that both optical and microphysical properties of smoke can vary significantly depending on its origin, persistence, and transport in the atmosphere. These factors that must be carefully incorporated into future climate models, especially considering the frequent occurrences of fire events worldwide. Full article

The continuous, effective operation of grid-connected photovoltaic (GCPV) systems depends on dependable inverter failure detection. Early, precise fault diagnosis improves general system dependability, lowers maintenance costs, and saves downtime. Although computing efficiency remains a difficulty, particularly in resource-limited contexts, machine learning-based fault detection presents interesting prospects in accuracy and responsiveness. By streamlining data complexity and allowing faster and more effective fault diagnosis, dimensionality reduction methods play vital role. Using dimensionality reduction and ML techniques, this work explores inverter fault detection in GCPV systems. Photovoltaic inverter operational data was normalized and preprocessed. In the next step, dimensionality reduction using Principal Component Analysis (PCA) and autoencoder-based feature extraction were explored. For ML training four classifiers which include Random Forest (RF), logistic regression (LR), decision tree (DT), and K-Nearest Neighbors (KNN) were used. Trained on the whole standardized dataset, the RF model routinely produced the greatest accuracy of 99.87%, so efficiently capturing complicated feature interactions but requiring large processing resources and time of 36.47sec. LR model showed reduction in accuracy, but very fast training time compared to other models. Further, PCA greatly lowered computing demands, especially improving inference speed for LR and KNN. High accuracy of 99.23% across all models was maintained by autoencoder-derived features. Full article