Search Results (784)

Under global warming conditions, assessing snow avalanche susceptibility and identifying critical influencing factors in the snow-covered regions of the Tibetan Plateau is fundamental to preventing snow avalanche disaster risks. This study focuses on the Eastern Himalayan Syntaxis (EHS) region and collected 449 snow avalanche events through event collection, field investigations, and remote sensing interpretation. Using these data, the study employed four machine learning models—Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and Gradient Boosting Decision Tree (GBDT)—to produce snow avalanche susceptibility maps. Furthermore, the SHAP algorithm was utilized to identify the critical factors affecting snow avalanche susceptibility in the EHS. Under the random split, the RF model achieved the highest AUC value (0.912). Under spatial block cross-validation, the AUC values decreased to varying degrees, with SVM and RF still performing best (AUC = 0.855 and 0.850, respectively). The total area of extremely high and high-risk snow avalanche zones was 4036.4 km², accounting for 21.0% of the study area. High-risk snow avalanche areas were primarily concentrated in the narrow valley regions near the entrances of the Doxiongla and Galongla tunnels. Additionally, the use of the SHAP algorithm enhanced the interpretability of the RF model. The study found that NDVI, glacial kernel density, January average temperature, land use, slope, and aspect were the main factors predicting snow avalanche occurrence, reaching a cumulative contribution rate of 66.4%. These findings not only confirm the effectiveness of the combination of machine learning models and the SHAP algorithm in snow avalanche susceptibility assessment but also delineate critical areas for enhancing local avalanche disaster prevention and mitigation strategies. The results of this study can serve as a reference for other mountainous regions with insufficient snow avalanche research data. Full article

This paper provides the design and simulation of a lightweight cryptographic protocol on smart house IoT devices using ChaCha20 and Ascon-AEAD128. The protocol, implemented in Python 3.14.6 and tested on a virtual ESP32 platform using Wokwi and CloudAMQP, uses stream cipher encryption with authenticated message tagging to provide confidentiality and integrity. Two conditions, normal and tampered transmission, were experimented to confirm tag validation and successful decryption. Findings affirmed sound detection of tampering and unauthorized access prevention, proving usefulness of current AEAD ciphers on limited devices. The protocol highlights cryptographic systems that prioritize computationally efficiency and robustness, which is essential in smart homes that have limited power and memory. The hybrid design provides confidentiality and authenticity using a minimal overhead by utilizing ChaCha20 to provide lightweight encryption and Ascon-AEAD128 as authentication. The resilience to the replay and modification attacks was demonstrated in experiments based on message queues and injected packet modifications simulating real-world conditions. Even though benchmarking of hardware was not carried out, the simulated values reveal stability and flexibility to use in low-power systems. This article emphasizes the necessity of authenticated encryption as default, which is consistent with the NIST standards and reflects the appropriateness of Ascon to new IoT security requirements. It also creates a reconfigurable structure that can be used in other highly constrained systems, such as healthcare monitoring and industrial IoT. The main contribution is the gap between theoretical cryptography and practical IoT security provided by the practical prototype. Further development will include tests on physical ESP32 modules and fine performance profiling, yet already, the current implementation proves a scalable, secure model of smart home IoT. Finally, this research demonstrates that the demand of reliable and low-power-based communication in resource constrained networks can be met efficiently without sacrificing device performance by means of lightweight cryptography. In addition to the cryptographic protocol design, this study integrates a machine learning-based intrusion detection layer trained on the Edge-IIoTset dataset, in which Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting classifiers are evaluated to complement the encryption–authentication framework with anomaly-aware monitoring of network traffic. Full article

This study presents a leakage-proof hybrid machine learning framework for early flood risk forecasting using multimodal spatiotemporal tabular data. An event-based dataset covering Kazakhstan from 2001 to 2021 was created by integrating topographic characteristics, hydrometeorological variables, long-term surface water dynamics, and remotely sensed spectral indices. To ensure realistic assessment, only pre-event observations were used, and event-based temporal data separation was employed to prevent leakage between the training and test subsets. The proposed Remote Sensing Adaptive Linear Opinion Pool Machine Learning (RS-ALOP-ML) framework combines multiple logistic regression experts with different regularization strengths through an Adaptive Linear Opinion Pool (ALOP) probabilistic fusion strategy, thereby preserving interpretability while improving forecasting robustness. The proposed framework was evaluated for four independent forecast horizons (T + 1, T + 7, T + 14, and T + 30 days) and compared with traditional machine learning algorithms and state-of-the-art tabular deep learning models, including Random Forest, ExtraTrees, XGBoost, LightGBM, CatBoost, MLP, FT-Transformer, TabNet, and Process Wide&Deep. Experimental results demonstrate that the proposed hybrid approach consistently achieves competitive or superior forecasting performance while maintaining computational efficiency and transparent probabilistic outputs. The study highlights that carefully designed hybrid machine learning architectures, combined with leakage-safe evaluation protocols, provide a robust foundation for multimodal environmental forecasting and decision support applications. Full article

Wildfire monitoring using synthetic aperture radar (SAR) provides critical capabilities under challenging atmospheric conditions where optical sensors are limited by smoke and cloud cover. We evaluated Sentinel-1 C-band SAR interferometric coherence for Burned-area detection of the 2017 “Las Máquinas” megafire (Maule, Chile), comparing Ascending (Asc) and Descending (Dsc) orbital geometries processed with the AMSTer InSAR software. Multi-temporal RGB Coherent Change Detection composites were constructed using two interferometric pairs per orbit: the Normalised Differential Activity Index (NDAI, R channel), pre-fire coherence (G channel), and co-event coherence (B channel), clearly delineating the fire scar through red and orange signatures reflecting fire-induced vegetation loss and soil exposure. Seven machine-learning classifiers (Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), K-Nearest Neighbours (KNN), Gradient Boosting Classifier (GBC), and XGBoost) were trained on the three-band coherence feature space. For the Ascending orbit, XGBoost achieved the highest performance (OA = 0.9328; F1 = 0.9195) and mapped 143,950 ha (76.2%) as Burned. For the Descending orbit, XGBoost also performed best (OA = 0.9221; F1 = 0.9055) and mapped 144,475 ha (76.5%) as Burned. In this case study, the Ascending geometry performed marginally better than the Descending one; however, the leading classifiers were statistically indistinguishable, indicating that the Burned and Unburned classes are close to linearly separable in the coherence feature space. These results confirm the effectiveness of coherence-based SAR analysis for large-scale wildfire mapping under adverse atmospheric conditions. Full article

Red mud, an alkaline industrial by-product of alumina refining generated in enormous volumes worldwide, poses a persistent environmental disposal challenge; using it as a partial cement replacement offers a promising route toward more sustainable concrete, but the resulting compressive strength is governed by complex, nonlinear interactions among the mix constituents that conventional empirical and regression-based models struggle to capture accurately. To address this challenge, the present study develops and compares four machine learning and deep learning models, namely the Deep Gradient Boosting Machine (DGBM), the Differentiable Neural Decision Tree (DNDT), Long Short-Term Memory (LSTM), and the Monte Carlo Dropout Neural Network (MCDNN), for the accurate and uncertainty-aware prediction of the compressive strength of red mud concrete. A dataset of 183 data points, compiled from the literature and supplemented with experimental results, was used to capture the influence of red mud content, curing period, and other mix parameters, including cement dosage, water content, and admixture proportions. The data were pre-processed prior to model training, and predictive performance was evaluated using R², RMSE, MAE, and WMAPE, among other indicators. The results show that the deep learning models outperformed the tree-based models: LSTM achieved the highest accuracy (R² = 0.942 on the testing dataset), while MCDNN additionally provided reliable uncertainty estimates alongside comparable prediction accuracy; DNDT and DGBM were comparatively less effective. Global sensitivity analysis identified fly ash and water content as the most influential contributors to strength development. By combining rigorous data-driven modeling with sensitivity and uncertainty analysis, this study contributes to the literature a validated, uncertainty-aware deep learning framework for sustainable concrete strength prediction, and offers practical value to the construction industry by providing engineers with a reliable, data-driven tool for optimizing red mud content in concrete mix design, thereby supporting the safe and wider industrial utilization of this problematic waste stream. Full article

Electric vehicle (EV) adoption is accelerating rapidly, increasing pressure on public charging infrastructure, urban energy systems, and network-expansion planning. However, many existing studies examine charging demand forecasting, station network characteristics, or infrastructure siting as separate problems, limiting their ability to translate observed charging behaviour into coordinated planning decisions. To address this gap, this study proposes a multi-domain machine learning framework that integrates spatio-temporal charging demand, station-level infrastructure characteristics, land-use information, and graph-based network relationships for EV charging prediction and infrastructure siting. The analysis uses 8,544,695 public charging transactions recorded at 8553 stations across Beijing during January and July 2025. Charging sessions are aggregated into an hourly station panel and modelled using seasonal-naïve and historical mean baselines, ridge regression, gradient-boosted trees, graph-augmented gradient boosting, and a gated recurrent unit network. Model reliability is assessed through temporal validation, feature ablation, spatial holdout testing, cross-season transfer analysis, explainability, and non-parametric statistical comparison. The results show that graph-augmented gradient boosting achieved the strongest RMSE and variance-explained performance, with an RMSE of 53.80 kWh and $R^2=0.734$, while the gated recurrent unit produced the lowest MAE of 23.82 kWh. Graph neighbour features yielded only a marginal and statistically non-significant forecasting improvement, indicating that the station network is structurally informative but predictively redundant once temporal history is available. For infrastructure expansion, NSGA-II achieved the highest Pareto front hypervolume and identified a knee-point solution of 156 additional chargers, reducing unmet demand by 43.4%. These findings demonstrate that integrated forecasting, graph analysis, and multi-objective siting can provide more defensible and operationally relevant evidence for city-scale EV charging planning. Full article

Machine learning combined with multispectral remote sensing provides an efficient and non-destructive approach for monitoring fruit tree nutrient status. However, conventional machine learning methods have limited ability to capture complex spectral-feature interactions, while deep neural networks often suffer from overfitting under small-sample field conditions. This study developed a UAV-based multispectral framework for estimating leaf nitrogen (N), phosphorus (P), and potassium (K) contents in a commercial pear orchard. Five-band spectral reflectance data were used to generate 21 vegetation indices (VIs), and Pearson correlation analysis and recursive feature elimination (RFE) were applied for feature optimization. A Deep Forest (DF)-based framework and a gradient boosting enhanced variant (GB-DF) were developed and compared with convolutional neural networks (CNN), support vector regression (SVR), random forest (RF), and gradient boosting decision tree (GBDT). Results showed that GB-DF achieved the best performance for N, P, and K estimation, with R² values of 0.6959, 0.7535, and 0.7216, and RMSE values of 0.7877, 0.1280, and 0.6059 g/kg, respectively. Moreover, GB-DF required only 6.35 s for training, being approximately twice as fast as GBDT and 5.7 times faster than CNN. The proposed GB-DF framework shows potential to improve nutrient estimation accuracy under limited-sample conditions and provides an effective solution for UAV-based nutrient mapping and variable-rate fertilization in precision orchards. Full article

Assessing visual fatigue is crucial in eye-controlled interaction. Traditional methods are either overly subjective or rely on highly invasive, costly equipment and complex procedures that require expert supervision. This study proposes a machine-learning-based approach for visual fatigue assessment. Data collection employs non-intrusive, easily monitored eye-tracking to capture ocular eye movement data and task performance data, while subjective questionnaires label fatigue states. For feature selection, participant-level Wilcoxon signed-rank tests with Benjamini–Hochberg FDR correction were used to identify fatigue-related indicators, and a redundancy-removal step based on Spearman correlation yielded a final set of six non-redundant features. For the assessment method, we introduced a gradient boosting decision tree–logistic regression (GBDT-LR) model whose hyperparameters are optimized via Bayesian optimization. All models were evaluated under a unified 5-fold stratified cross-validation framework with within-fold standardization and nested hyperparameter tuning. Results indicate that this model can effectively predict the state of visual fatigue. Compared with the performance of five other models—gradient boosting decision tree (GBDT), logistic regression (LR), support vector machine (SVM), random forest (RF), and RF-SVM—the proposed GBDT-LR model achieved an assessment accuracy of 89.79%, demonstrating strong predictive performance. This study provides an effective method for predicting visual fatigue in eye-controlled interaction, laying a research foundation for optimizing the user experience of eye-controlled interaction and promoting the sustainable development of eye-control technology. Full article

The Coordinate Rotation Digital Computer (CORDIC) algorithm is widely employed in digital signal processing and hardware accelerators because it computes a broad range of elementary functions using iterative shift-and-add operations. Conventional CORDIC implementations, however, execute a fixed number of iterations irrespective of the input characteristics or the precision required, resulting in unnecessary computational overhead and increased execution latency. This work presents an explainable machine learning framework for adaptive iteration optimization in a multi-mode CORDIC architecture supporting circular, hyperbolic, and linear operating modes. A unified prediction framework for calculating the optimal number of iterations is made possible by the developing a generic feature representation to describe the numerical behavior of CORDIC computations across various modes. We systematically evaluated eight regression models, including Linear Regression, Decision Tree, Random Forest, Extra Trees, Support Vector Regression, Multi-Layer Perceptron, and Extreme Gradient Boosting (XGBoost) and LightGBM. Among the models evaluated, the Decision Tree achieved the best performance on an independent test set of 2305 samples from 461 previously unseen input groups, with a MAE of 0.9160 iterations, RMSE of 1.9671, and $R^2$ of 0.6076. Predictions were within one and two iterations of the reference value for 80.26% and 90.07% of the test samples, respectively. Since prediction accuracy alone does not guarantee that the required numerical tolerance will be satisfied, the predicted iteration count was further evaluated using the actual CORDIC error, followed by a safety-correction procedure. The safety-corrected approach achieved 100% tolerance satisfaction on the independent test set, reducing the mean number of iterations from 20 to 11.739, corresponding to a 41.31% reduction in iterations. Model behavior was further interpreted using feature importance analysis, permutation importance, and feature ablation studies to examine the contribution of individual features to iteration prediction. Statistical robustness is established using bootstrap confidence intervals, the Friedman test, and Holm-corrected Wilcoxon signed-rank tests. Full article

Hearing aid technology adaptation and user satisfaction are influenced by multiple interacting demographic, clinical, and behavioral factors, making reliable prediction of outcomes challenging with conventional statistical approaches alone. This study proposes an explainable machine learning framework to investigate the multidimensional determinants of hearing aid satisfaction by integrating demographic characteristics, hearing aid-related variables, and patient-reported outcomes obtained from the Hearing Aid Technology Adaptation and Satisfaction Scale (HATASS). Five regression algorithms—Linear Regression, Decision Tree Regression (DTR), Random Forest Regression (RFR), Support Vector Regression, and Gradient Boosting Regression (GBR)—were comparatively evaluated using a five-fold cross-validation strategy. Predictive performance was assessed using the root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination ($R^2$), while model interpretability was investigated through cross-validated out-of-bag permutation feature importance analysis. Among the evaluated algorithms, Random Forest Regression achieved the most consistent predictive performance, yielding the lowest average RMSE (14.225) and the highest average $R^2$ (0.146) under the adopted validation framework. Although the overall predictive performance remained modest, the explainability analysis consistently identified age as the most influential predictor, followed by onset year, education level, hearing aid usage duration, and daily hearing aid use. In contrast, gender, battery type, tinnitus, and vertigo contributed comparatively less to model predictions. These findings indicate that hearing aid adaptation and satisfaction arise from complex nonlinear interactions among demographic, behavioral, clinical, and device-related characteristics rather than isolated linear associations. The proposed framework provides an interpretable, internally validated analytical approach for investigating hearing aid technology adaptation and satisfaction and establishes a foundation for future studies that integrate comprehensive audiological measurements, longitudinal follow-up data, and independent external validation to support the development of more transparent and personalized hearing healthcare systems. Full article

Background/Objectives: Dental age estimation is important in clinical and forensic practice, particularly when skeletal indicators are unavailable or compromised. Machine-learning models often achieve lower regression errors than conventional dental methods; however, whether this translates into better classification performance at legally relevant age thresholds remains unclear. This study compared the Demirjian and Willems methods with several machine learning models for overall accuracy and threshold-specific performance at the jurisdiction-specific ages of 12 and 15 years. Methods: A total of 1384 panoramic radiographs from individuals aged 8.00–15.99 years were retrospectively evaluated. The developmental stages of the seven left mandibular permanent teeth and sex were used as model inputs. Linear Regression, Decision Tree, Random Forest, Support Vector Regression, Multilayer Perceptron, Gradient Boosting, and XGBoost were trained using cross-validation and evaluated on an internal holdout set. Performance was assessed using regression errors, age-group-specific bias, sensitivity, specificity, balanced accuracy, and likelihood ratios. Results: Machine-learning models generally produced lower errors than conventional methods. In the holdout set, the lowest mean absolute error was 0.512 years for Support Vector Regression and Gradient Boosting, followed by 0.519 years for XGBoost, compared with 0.649 and 0.654 years for the Willems and Demirjian methods. However, lower regression error did not consistently improve threshold-specific performance. At 12 years, machine learning models increased sensitivity but reduced specificity and positive likelihood ratios relative to Willems. At 15 years, Linear Regression and Random Forest produced no positive predictions, whereas the better-performing models showed results similar to Willems. Conclusions: Lower regression error does not necessarily indicate better forensic threshold-specific classification performance. Dental age-estimation models should therefore be validated using threshold-specific likelihood ratios, classification metrics, and age-group-specific bias in addition to overall prediction errors. Full article

Objective: This study aimed to develop machine learning (ML)-based models for the early prediction of macrosomia using only maternal sociodemographic and obstetric data. Methods: This retrospective study included 100 pregnant women who delivered at the Obstetric Clinic of Eskisehir City Hospital between January 2022 and December 2023. Participants were classified as nulliparous (n = 48) or parous (n = 52) and further categorized according to the presence or absence of fetal macrosomia (birth weight > 4000 g). Predictor variables included maternal age, body mass index (BMI), gravida, smoking status, history of diabetes mellitus, and hypertension; previous birth weight was additionally included in the parous model. Separate machine learning models were developed for nulliparous and parous women using Extra Trees Classifier, Light Gradient Boosting Machine (LGBM), eXtreme Gradient Boosting (XGB) Classifier, Random Forest, and Logistic Regression. The dataset was randomly divided into training (80%) and testing (20%) subsets. Internal validation was performed using 10-fold cross-validation within the training dataset to optimize model performance and reduce overfitting. Given the relatively small sample size, this study was designed as a pilot exploratory investigation. Results: Overall, 48 nulliparous and 52 parous mothers were included in the study. Among the nulliparous women, 22 (45.8%) had macrosomic newborns, whereas 26 (54.2%) had normal birthweight newborns. Among parous women, 28 (53.8%) had macrosomic newborns, while 24 (46.2%) had normal birthweight newborns. For nulliparous mothers, the XGB Classifier achieved the highest accuracy (80%) and AUC-ROC (82.2%), demonstrating robust predictive performance. For parous mothers, the XGB Classifier again outperformed other ML models, achieving an accuracy of 72.7% and an AUC-ROC of 83.3% in predicting macrosomia. Conclusions: This study highlights the feasibility of ML-based decision support systems in obstetrics, particularly in low-resource settings, to predict macrosomia using readily available maternal characteristics. As a pilot study, these findings should be interpreted cautiously and require validation in larger multicenter cohorts before clinical implementation. Full article

The nature of solar radiation and the high penetration of photovoltaic (PV) systems in the smart electrical grid necessitate the development of models driven by historical operational data capable of precisely estimating the performance of PV systems. In this work, six proposed models are applied to predict the conversion efficiency of a 7.98 kWp rooftop on-grid PV system in Jordan. The dataset comprises 179 daily samples obtained during a single spring–summer period (17 March–24 September 2014). The efficiency modeled is the combined efficiency of the modules and inverter as a system. The proposed models are Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting (GB), Gaussian Process Regression (GPR), and Elastic Net (EN). The effectiveness of these proposed models is assessed by calculating four performance metrics, namely, the Mean Square Error, prediction accuracy, Coefficient of Determination (R²), and adjusted R², and benchmarking the results with those of six prediction models discussed in our previous work. The results from the unweighted Decision-Making Matrix show that RF showed the best overall performance among the proposed and benchmark models considered. By contrast, the SVM, DT, and GB models exhibited moderate predictive behavior. However, Elastic Net is the worst-performing model among the 12 proposed and benchmark models discussed in this work. Moreover, the RF model’s consistently low prediction error supports its practical utility for PV system performance estimation, despite a slightly higher training cost than simpler models. Full article

Accurate spatial information on soil pH is essential for soil management, agricultural decision-making, and ecosystem assessment. Although Earth observation (EO) data have played an increasingly important role in digital soil mapping (DSM), most studies have relied mainly on optical imagery, while synthetic aperture radar (SAR) information, especially interferometric coherence, remains underutilized for soil pH prediction. This study explored the value of Sentinel-1-derived interferometric coherence and backscatter images, Sentinel-2 optical imagery, and topographic–climatic variables for national-scale mapping of soil pH across Spain. Models were developed using random forest (RF) and boosted regression trees (BRT) with 3867 LUCAS 2018 topsoil samples under 11 prediction scenarios representing different radar configurations, radar-derived feature types, and multi-source data integration strategies. VH backscatter performed better than VV backscatter, while combining backscatter from both polarizations and both orbit directions further improved performance within the backscatter-only group. When different predictor groups were used separately, coherence images achieved R² values of 0.49–0.52, outperforming all other individual predictor groups. Under BRT, adding coherence to backscatter increased R² from 0.45 to 0.56 for pH in CaCl₂ and from 0.46 to 0.57 for pH in H₂O, and the further inclusion of Sentinel-2 optical imagery slightly improved performance. The best performance was achieved by integrating all satellite-derived variables with topographic and climatic predictors, with R² values of 0.62 for both pH in CaCl₂ and pH in H₂O under BRT. Variable importance analysis further identified coherence as the most influential predictor group within the evaluated predictor set, with short-temporal-baseline coherence features ranking highest. The predicted maps revealed clear spatial heterogeneity, with lower pH values mainly in northern and northwestern Spain and higher values in central and southeastern regions. These findings demonstrate the added value of Sentinel-1 interferometric coherence for national-scale soil pH mapping. Full article

The use of deepfake media, particularly face-swap and face-shifter videos, has proliferated rapidly on social media and online news outlets. Though there are legitimate uses for this synthetic media in certain applications, such as media reporting, there is always the possibility of misleading people, creating distrust of digital information, and even posing security threats. Most existing studies have focused on deepfake detection using either image or video frames; however, these methods are computationally costly and do not perform well in real time. To address these limitations, this work proposes a pipeline for deepfake video detection using network packet analysis. A novel PCAP dataset is constructed by streaming real and manipulated video content over WebRTC and TCP (Port 8080) protocols, and machine learning models are then trained on 48 extracted network-level features to distinguish deepfake traffic from authentic streams. For implementation, four deepfake video datasets of varying quality are utilized, namely, HIDF, FaceForensics++, SDFVD-V2, and ManualFake-2022, each containing both real and manipulated samples. Videos are segmented into 3, 6, and 9-s clips and sequentially streamed over WebRTC and TCP (Port 8080) protocols to capture network traffic in PCAP format. Experiments are conducted using six classical machine learning classifiers, namely KNN, Logistic Regression, Decision Tree, Random Forest, Histogram Gradient Boosting, and Naïve Bayes, trained on 48 network-level features. The proposed pipeline achieved the highest classification accuracy of 91.2% on TCP (Port 8080) and 79.9% on the WebRTC protocol, both obtained using 6-s video captures. Full article