ABSTRACT Understanding the optical properties of different sea ice types is crucial for improving knowledge of energy balance, heat and mass budgets, and for enhancing the accuracy of remote sensing retrievals. The in situ-based spectral characteristics of coastal first-year sea ice in eastern Liaodong Bay, Bohai Sea, were systematically analysed. A regional reflectance database was established in January 2025, collecting 49 ice samples with diverse types, thicknesses, densities, sampling locations, and observation angles. The results show that scale-like ice exhibited the highest visible-band reflectance (>0.7), followed by young ice with considerable variability, while frazil ice, nilas, and needle ice exhibited lower reflectance due to their thin structure, loose texture, and uneven surfaces, and sediment-laden ice had the lowest reflectance. Sea ice reflectance was strongly related to ice density, with lower-density ice showing higher values, and displayed a significant exponential relationship with ice thickness in both ultraviolet and visible bands (p < 0.01), with the ultraviolet band yielding higher coefficients of determination. The Hemispherical Directional Reflectance Factor (HDRF) increased with azimuth angle, reaching a minimum at nadir (0° zenith), and reflectance grew with zenith angle, with forward scattering being dominant, followed by side and back scattering. Additionally, comparisons between Geostationary Ocean Colour Imager II (GOCI-II) satellite data and in situ measurements showed similar spectral patterns, including primary peaks in the ultraviolet–blue bands and a secondary maximum near 600–700 nm, although GOCI-II reflectance was generally 10–20% lower. Moreover, preliminary results based on four machine learning algorithms indicate that GOCI-II satellite data have clear discriminatory potential for classifying sea ice by both region and thickness.

The rapid development of big data is profoundly reshaping the insurance industry, while traditional actuarial methods face obvious limitations in dealing with high-dimensional, nonlinear, and unstructured data. This paper systematically reviews the current application status of machine learning (ML) in the insurance industry under the background of big data, focusing on four key areas: fraud detection, risk assessment and management, claims issues, and customer segmentation. Integrated empirical research shows that ML methods generally outperform traditional statistical models in terms of prediction accuracy, operational efficiency, and decision support. Meanwhile, this paper identifies four types of real-world challenges that restrict the responsible deployment of machine learning: data quality issues, regulatory and ethical constraints, privacy and security concerns, as well as organizational and operational barriers. Based on the literature review, this paper holds that machine learning does not replace actuarial science, but rather substantially enhances its predictive ability, operational efficiency and strategic decision-making level. Responsible technology application must simultaneously promote the construction of transparency, fairness and compliance. This paper constructs an analytical framework that integrates technical capabilities with practical constraints, providing a structured reference for subsequent research and practical guidance for insurance companies and policymakers to promote the sustainable and compliant application of machine learning.

Conventional microbial diagnostic techniques encounter considerable obstacles, such as prolonged turnaround times, labor-intensive protocols, and constraints in precision. To fix these problems and improve diagnostic capabilities, clinical microbiology is using more and more artificial intelligence (AI) technologies. To systematically evaluate the present applications, developments, and influence of artificial intelligence technologies in clinical and diagnostic microbiology, emphasizing pathogen identification, antimicrobial resistance detection, and laboratory automation. A thorough systematic literature search was performed utilizing the PubMed, Scopus, Web of Science, and Google Scholar databases from 2020-2024. Search terms comprised combinations of “artificial intelligence”, “machine learning”, “clinical microbiology”, “diagnostic microbiology”, “pathogen identification”, and “antimicrobial resistance”. Studies detailing AI applications in clinical microbiology were included, whereas non-English articles and review papers were excluded. Eighty-nine studies met the requirements for inclusion. Machine learning algorithms showed high accuracy (85%-99%) in finding pathogens in different types of samples. Deep learning models outperformed others in predicting antimicrobial resistance, with AUROC (Area Under the Receiver Operating Characteristic) values above 0.83. AI-enhanced microscopy and automated image analysis cut down on the time it took to make a diagnosis from days to hours while keeping the sensitivity (92%-98%) and specificity (81%-95%) high. AI technologies have transformed clinical microbiology by delivering swift and precise diagnostic solutions. Combining machine learning with MALDI-TOF MS (Matrix-Assisted Laser Desorption/Ionization Time-of-Flight), automated microscopy, and genomic analysis has made it easier to find pathogens and test for antibiotic resistance. AI is a game-changing force in modern diagnostic microbiology, even though it is hard to standardize and use.

Abstract Purpose To map and synthesise current evidence on machine learning (ML) applications for anterior cruciate ligament (ACL) injury risk estimation, rehabilitation monitoring and return‐to‐sport (RTS) decision support, with emphasis on clinical relevance and methodological quality. Study Design Scoping review with descriptive evidence synthesis. Methods The review was conducted in accordance with PRISMA‐ScR guidelines. Four electronic databases ( n = 4) (PubMed, Scopus, IEEE Xplore and Web of Science) were searched for peer‐reviewed studies published between 2016 and 2025. Eligible studies applied ML models to ACL injury prediction, postoperative recovery assessment, or RTS evaluation. Data were extracted on study design, participant characteristics, data modalities, ML algorithms and clinical endpoints. Reporting quality, risk of bias and certainty of evidence were assessed using TRIPOD‐AI, PROBAST‐AI and an adapted GRADE framework. Quantitative results were summarised descriptively rather than pooled meta‐analytically. Results Forty studies ( n = 40) met the inclusion criteria. Tree‐based ensemble models, particularly Random Forest and Extreme Gradient Boosting, were most frequently applied and showed consistent performance across clinical, biomechanical and wearable datasets. Deep learning models were predominantly used for imaging‐based tasks such as ACL tear detection and graft assessment. Wearable and sensor‐integrated approaches supported continuous functional monitoring and RTS readiness estimation. Methodological quality was generally acceptable, although external validation and standardised outcome definitions were inconsistently reported. Conclusion ML approaches demonstrate growing potential as adjunctive clinical decision‐support tools in ACL rehabilitation and RTS assessment. Wider clinical adoption will require standardised multimodal datasets, external validation and explainable modelling to ensure safe, interpretable and context‐appropriate implementation. Level of Evidence Level II, high‐quality scoping review with structured synthesis of cohort‐based prognostic and predictive modelling studies.

Hypertrophic cardiomyopathy (HCM) is the most common inherited cardiovascular disease. Recent metabolomics studies have revealed pathogenic mechanisms and provided new perspectives for diagnosis. This study aimed to analyze plasma metabolic alterations and construct a preliminary diagnostic model for HCM based on untargeted metabolomics and machine learning (ML) algorithms, in order to explore potential pathogenic pathways and improve diagnostic accuracy during screening. A total of 76 HCM patients and 35 normal participants were consecutively recruited from August, 2023 to December, 2023. Data were split into discovery and validation sets at a ratio of 7:3 and the feature combinations were selected using support vector machine (SVM) and random forest (RF). Stepwise multivariate linear regression analysis was performed to identify key metabolites associated with left ventricular wall thickness. Metabolic pathway analysis was performed using KEGG. Totally 1481 metabolites were identified with 640 differential metabolites and 240 significant differential metabolites. Multivariate statistical analysis showed that metabolism results could effectively differentiate the two cohorts (OPLS-DA positive ion mode R2Y = 0.744, Q2 = 0.456; negative ion mode R2Y = 0.611, Q2 = 0.441). SVM and RF screened the same combination of features including 7-keto-8-aminopelargonic acid (KAPA), γ-linolenoyl ethanolamid, nitrilotriacetic acid, D-quinovose and N-acetyl-l-aspartic acid (NAA), which could effectively and accurately differentiate HCM patients from normal participants (in discovery and validation sets, the SVM model AUROC was 0.996 and 0.985 with accuracies of 96.1% and 97.1%, respectively; the RF model AUROC was 1.000 with accuracies of 94.8% and 100.0%, respectively). In metabolic pathway analysis, central carbon metabolism in cancer and protein digestion and absorption were significantly upregulated in HCM patients, which were connected by alanine, aspartate and glutamate metabolism. Stepwise multivariate linear regression analysis revealed that NAA was correlated with left ventricular mass index and RV5+SV1 (P < 0.05), which may be the central target of the connecting pathway. Plasma metabolite diagnostic model including KAPA, γ-linolenoyl ethanolamid, nitrilotriacetic acid, D-quinovose and NAA can effectively and accurately screen HCM patients. Metabolomics combined with ML algorithm showed that alanine, aspartate and glutamate metabolism may be the pathogenic pathway leading to the occurrence of HCM with NAA as the central target.

Sentiment classification plays a crucial role in analyzing customer feedback to identify market trends, enhance product recommendations, and improve customer satisfaction. This study focuses on sentiment analysis of Amazon reviews using two major datasets—Fine Food Reviews and Unlocked Mobile Reviews—which exhibit label imbalance. To address this challenge, both oversampling and undersampling techniques were applied to balance the datasets. Various machine learning (ML) algorithms, including Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Naïve Bayes (NB), and Gradient Boosting Machine (GBM), as well as deep learning (DL) models such as Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and transformer-based models like RoBERTa, were implemented. After data cleaning and preprocessing, models were trained, and performance was evaluated. The results indicate that oversampling significantly enhances classification accuracy, particularly for the Fine Food dataset. Among ML models, Random Forest achieved the highest accuracy due to its ensemble approach and robustness in handling high-dimensional data. DL models, particularly RoBERTa, also demonstrated superior performance owing to their capacity to capture contextual dependencies. The findings emphasize the importance of data balancing for optimal sentiment analysis and contribute valuable insights toward advancing automated opinion classification in e-commerce applications.

This study aimed to develop and externally validate machine learning (ML)-based models to characterize surgical classification patterns between hysterectomy and myomectomy using fibroid characteristics and female sex hormone profiles. This multicenter study included 600 women with uterine fibroids (UFs) who presented to 3 hospitals. Of these, 362 (60.3%) underwent hysterectomy, while 238 (39.7%) underwent myomectomy. Statistical analyses and ML models were applied to both groups. ML model development was performed using individual and combined inputs of female sex hormones together with fibroid characteristics. Five ML classification algorithms were evaluated, including support vector machines, decision trees, random forests, k-nearest neighbors, and logistic regression. In total, 2555 model-input combinations were tested. The performance of the selected best-performing model was further evaluated using an independent, blinded external validation cohort comprising 30 cases. Women in the hysterectomy group had significantly higher mean age, follicle-stimulating hormone, luteinizing hormone, UF number, UF volume, uterine volume, disease duration, gravidity, parity, and prolactin (PRL) levels compared with the myomectomy group (all P < .001). In contrast, estradiol and anti-Müllerian hormone levels were significantly lower in the hysterectomy group (P < .001). Across all modeling experiments, 2012 of 2555 model-input combinations achieved perfect classification performance (accuracy = 100%) when sex hormone profiles and UF characteristics were jointly used as inputs. Models using UF number alone also demonstrated high predictive performance, with accuracy reaching up to 96%. Agreement between algorithmic predictions and final surgical decisions was observed in 97% of cases, with one discordant case identified at a clinically borderline threshold. ML models trained on hormone profiles and fibroid characteristics were able to reproduce prevailing surgical classification patterns, largely reflecting strong baseline separability driven by age- and menopause-associated hormonal profiles, with consistent performance observed in an independent blinded validation cohort. These findings support the feasibility of quantitatively modeling routine decision structures, while highlighting the need for further validation in clinically heterogeneous and ambiguous cases.

ABSTRACT Air pollution poses a significant threat to skin health, contributing to inflammation, aging, and disruption of the epidermal barrier. This study aims to identify genes associated with skin exposure to air pollution and skin barrier damage, and to discover potential biomarkers for these effects. Datasets related to air pollution‐exposed skin (PS) and skin barrier damage (SD) were obtained from the Gene Expression Omnibus (GEO) database. GO and KEGG enrichment analyses were initially performed on both datasets. Through functional enrichment analysis, protein–protein interaction (PPI) network construction, and the application of two machine learning algorithms, we identified 140 common genes and two key diagnostic genes, FOSL1 and TKT. Receiver operating characteristic (ROC) curve analysis was used to validate the PS and SD datasets, achieving optimal area under the curve (AUC) values. Further investigation of FOSL1 and TKT via Gene Set Enrichment Analysis (GSEA) and immune cell infiltration analysis explored their roles in PS and SD conditions. Additionally, a skin model simulating air pollution exposure using particulate matters (PMs) was developed. RT‐qPCR results showed that as the concentration of PMs increased, genes related to skin barrier damage were activated. The reliability of FOSL1 and TKT was confirmed through both RT‐qPCR and Western Blot analyses.

The increasing use of artificial intelligence (AI) in smart environments, industrial automation, healthcare monitoring, and Internet of Things (IoT) networks has increased the computational requirements and power consumption of cloud computing infrastructure. The use of cloud computing for AI processing faces challenges such as high latency, high bandwidth consumption, privacy concerns, and environmental emissions due to the large number of data centers. This work introduces a sustainable and low-power AI model based on the smart edge architecture․ It includes energy-efficient machine learning algorithms deployed at the edge for making smart decisions․ This model achieves optimal performance using lightweight model deployment, dynamic resource allocation, and hardware-aware optimizations such as model quantization and hardware pruning techniques․ A modular architecture is proposed that increases efficiency by focusing on data acquisition, edge processing, AI inference, energy management, and cloud synchronization․ Performance metrics include latency, energy consumption, compute efficiency, and inference quality․ The latter, in particular, is important for providing efficient AI capabilities on devices with limited resources․ Results of experiments show that local processing reduces network and energy use compared to processing in the cloud․ The proposed framework enables scalable and efficient deployment of AI while minimizing the environmental impact and maintaining performance to support sustainable computing․

The growing frequency and extent of wildfires constitute a significant environmental challenge, posing serious threats to ecosystems, biodiversity, and human livelihoods. This study presents a comprehensive wildfire susceptibility assessment for El Tarf Province, one of the most fire-prone yet understudied regions in Algeria. Long-term Landsat imagery (1995-2024) combined with four machine learning algorithms was used to produce high-resolution susceptibility maps and identify the key environmental and bioclimatic drivers of wildfire occurrence. Ten conditioning factors representing topographic, vegetative, edaphic, and climatic conditions were integrated, with elevation, Enhanced Vegetation Index (EVI), wind speed, and precipitation emerging as dominant predictors. Among the tested models, Random Forest achieved the highest predictive performance (ROC-AUC = 0.897), closely followed by XGBoost (0.896), while LightGBM provided an optimal balance between accuracy (0.875) and computational efficiency. Logistic Regression, though simpler, performed reasonably well (0.794). The Landsat-derived wildfire inventory comprised approximately 622,221 burned pixels and was subsequently split into a pre-2017 training set (72.8%) and a post-2017 testing set (27.2%) to evaluate model generalization over time. Spatial block cross-validation was applied to reduce spatial autocorrelation and enhance model generalization. This methodological framework, combining spatial and temporal validation, temporal hold-out, and spatial blocking, strengthens the robustness and reliability of wildfire susceptibility modeling. Interpretability analyses based on SHAP values, Gini importance, and permutation importance identified the contributions of underexplored variables, including vegetation type, soil type, and soil organic carbon (SOC). The resulting susceptibility maps provide valuable insights for spatial planning and ecosystem management, supporting evidence-based strategies to enhance environmental resilience and biodiversity conservation in Mediterranean landscapes.

Objective To mitigate current research limitations, this cross-sectional study aimed to systematically evaluate the associations between dietary amino acids and overweight/obesity and to identify critical biomarkers among Chinese children and adolescents. This was achieved by integrating multiple machine learning algorithms with traditional statistical models. Methods This study utilized data from the 2016–2019 China Children and Lactating Women Nutrition and Health Surveillance, a nationally representative survey. Participants included children and adolescents aged 6–18 years. Dietary intake was assessed using a validated food frequency questionnaire, and amino acid intakes were calculated. Four machine learning algorithms were applied to build prediction models. Model performance was evaluated via the area under the receiver operating characteristic curve (AUC). The SHapley Additive exPlanations (SHAP) method was used to interpret the optimal model and identify important features. Multivariable logistic regression models were additionally used to examine the relationship between amino acids and overweight/obesity risk. Results A total of 8,664 participants were included. The LightGBM model showed the best predictive effect (AUC = 0.805). Both SHAP analysis and logistic regression results consistently identified leucine (OR 1.13; 95% CI 1.01 ~ 1.27), threonine (OR 1.41; 95% CI 1.22 ~ 1.63), methionine (OR 1.30; 95% CI 1.07 ~ 1.57), and cysteine (OR 0.71; 95% CI 0.59 ~ 0.84) as key amino acids associated with overweight/obesity risk. After multivariable adjustment, the intake of leucine, threonine, and methionine was positively related to the risk of overweight/obesity, whereas cysteine intake was inversely related to the risk. Restricted cubic spline analyses suggested linear relationships for these associations. Conclusion Higher dietary intakes of leucine, threonine, and methionine are potential risk factors, while cysteine is a potential protective factor against overweight/obesity in Chinese children and adolescents.

The emergence of quantum computing poses a significant threat to the security of traditional encryption methods employed in satellite communication. To mitigate this vulnerability and enhance cybersecurity in the next generation of communication systems, a novel physical-layer solution is presented. This approach centers on enhancing satellite link security through the analysis of stochastic atmospheric scintillation, facilitated by machine learning (ML). The proposed method safeguards ground stations against Machine-in-the-Middle (MITM) attacks perpetrated from aerial platforms (AP) such as drones or Unmanned Aerial Vehicles (UAVs). The underlying principle leverages the distinct statistical parameters inherent to received signals. These parameters are contingent upon the specific propagation channel, which is influenced by rapid tropospheric scintillation. As signals from legitimate satellites and malicious drones traverse separate spatial paths within the dynamic atmosphere, they exhibit demonstrably divergent scintillation statistics. Wavelet filtering is employed to extract these statistics from the incoming signal. The extracted data is subsequently processed through an ML algorithm, enabling the differentiation between satellite signals and potential spoofing signals emanating from drones. Extensive simulations have been conducted, illustrating the efficacy and robustness of the proposed architecture, consistently achieving an authentication rate exceeding 98% across diverse scenarios. Additionally, experimental results obtained from measurement data collected from Nilesat and Eutelsat satellites at a ground station in Israel provide empirical validation for this innovative approach.

Machine learning (ML) accelerates progress in many areas, including biomedical and clinical research. ML algorithms provide powerful options for efficiently analyzing multivariate data sets. We developed and validated an ML pipeline to detect myelodysplastic syndrome (MDS)-associated pathological alterations of extracellular matrices (ECMs) by time-of-flight secondary ion mass spectrometry (ToF-SIMS). A Bayesian-optimized neural network (NN) was trained and applied to classify ToF-SIMS spectra of ECM secreted by mesenchymal stromal cells (MSCs) derived from MDS patients and healthy reference donors. Validated by principal component analysis, the explainer tool SHapley Additive exPlanations (known as SHAP) was integrated into the analysis pipeline to unravel characteristic compositional and structural differences of the ECM variants. Our results demonstrate the potential of ToF-SIMS-ML for the label-free investigation of pathogenic alterations of the ECM. Integrated into the multiscale ECM analysis of cell and organoid-based disease models, the introduced methodology may facilitate advances in the development of novel diagnostic and therapeutic strategies.

Ophiocordyceps sinensis (Berk.) is a functional food with health. O. sinensis quality varies by geographical origins, and current identification methods are sophisticated and time-consuming. This study aims to develop a rapid and straightforward method for accurately identifying O. sinensis geographic origins. Surface-enhanced Raman spectroscopy (SERS) was applied to analyze O. sinensis from four major production areas in China. Liquid chromatography-mass spectrometry (LC-MS) was used as a reference method to characterize compositional differences among samples and to verify the geographical authenticity of O. sinensis from the four production areas. Six machine learning (ML) algorithms were introduced for predicting geographical origins, and evaluation metrics were used to assess model performance. According to the comparative analysis, the Support Vector Machine (SVM) model performed best with the highest discrimination accuracy. A feature importance map was constructed to understand further how the model makes predictive decisions, revealing the significant Raman shifts in classifying O. sinensis from different geographical origins. The SERS-SVM method developed in this study contributes to the authenticity identification of O. sinensis geographical origins. It shows the potential to serve as an effective quality control method for the valuable TCM (traditional Chinese medicine).

Abstract. This paper introduces the extreme theory of functional connections (X-TFC), a physics-informed machine learning algorithm, and tailors it to estimate the remaining useful life (RUL) of wind turbine gearbox bearings experiencing fatigue crack growth. Unlike purely data-driven methods, X-TFC embeds a physics model, based on Head's theory in this work, into its training objective. The core of X-TFC is a random-projection single-layer neural network trained via an extreme learning machine, which requires only limited damage progression data and solves for output weights with a least-squares optimization algorithm. A composite loss function balances the network's fit to observed degradation data against the residuals of the governing crack growth differential equation, ensuring the learned damage trajectory remains physically plausible. When applied to a vibration-based health-index (HI) dataset measured during the growth of a crack on the inner ring of a high-speed bearing in a wind turbine gearbox (Bechhoefer and Dubé, 2020), X-TFC achieves near-zero prediction bias. Even when trained on only the first 10 %–20 % of the damage progression data, with sufficient physics weighting its predictions remain monotonic and smooth, delivering high prognosability and trendability. To quantify the epistemic uncertainty, we employ a Monte Carlo ensemble of independently initialized X-TFC models trained on noise-perturbed data, which yields confidence intervals around each RUL estimate and captures both model-parameter and epistemic uncertainty. In addition to a vibration-based HI, we demonstrate that the proposed framework can be directly applied to a supervisory control and data acquisition (SCADA) data-based HI (Eftekhari Milani et al., 2026) measured during similar wind turbine gearbox bearing crack faults, preserving its accuracy and interpretability. This extension shows the versatility of our approach, which is applicable to bearings of multiple gearbox manufacturers, models, and ratings using only SCADA data. By integrating domain knowledge with machine learning, X-TFC offers a rapid, reliable tool for crack prognostics. Its adaptability to other bearing failure modes, such as pitch bearing ring cracks, positions X-TFC as a powerful enabler of data-driven, physics-informed asset management in the wind energy sector and beyond.

China’s vast rural landscape exhibits pronounced regional disparities in both foundational resources and development potential. In the context of nationwide rural revitalization efforts, the emergent divergence in village development pathways underscores a pressing need for context-specific, classified interventions. To furnish a scientifically grounded typology of villages and inform differentiated development planning, this investigation focuses on Hubei Province as an illustrative case. Synthesizing survey data from 32,457 villages, we developed a multidimensional evaluation framework encompassing four pivotal domains: economic vitality, social service provision, ecological integrity, and cultural value. Leveraging the Self-Organizing Feature Map (SOFM) neural network—an unsupervised machine learning algorithm—we performed a cluster analysis on multi-source, heterogeneous datasets. This technique enabled the objective delineation of spatial typological patterns among Hubei’s villages, elucidated their underlying classification architecture shaped by multifaceted drivers, and demonstrated the methodological robustness and applicability of this approach for large-scale village categorization. Grounded in the derived typologies and informed by strategic directives from higher-tier planning instruments, we conducted a nuanced examination of the distinctive attributes characterizing each village type. The findings provide scientific evidence and decision-making support for village classification and rural revitalization planning in Hubei Province, with valuable implications for other regions with similar development foundations in China.

Aim. The aim of this work is to improve the quality of multi‑class classification for Intrusion Detection Systems (IDS) in the Internet of Things (IoT) environment. The goal of the research is to determine the impact of preliminary binary traffic filtering and the application of ensemble models on prediction accuracy, especially for minority attack classes, taking into account the computational constraints of IoT environments. Methods. Three architectural approaches were studied: direct multi‑class classification, direct multi‑class classification (including the “normal” class), and a hierarchical architecture based on initial binary detection followed by classification by attack type. Eight machine learning algorithms, as well as three ensemble methods (Soft Voting Classifier (SVC), Hard Voting Classifier (HVC), and Stacking Classifier (SC)), were evaluated. Experiments were conducted on the UNSW‑NB15 dataset using metrics such as Precision, Recall, and F1‑score. Results. The results show that direct classification provides better overall attack coverage (average F1‑score up to 63% for Gradient Boosting Classifier(GBC)), but may require longer training times (over2000 seconds for GBC). Hierarchical binary filtering significantly reduces computation time but can decrease performance for some rare classes. The GBC, Random Forest (RF), and Extra Trees (ET) algorithms stand out for their performance. Among the ensemble methods, the Stacking Classifier (SC) demonstrates the best results (F1‑score of 73.87%), surpassing individual classifiers, although it also requires substantial training time. Conclusion. This research shows that implementing binary filtration is a relevant strategy for reducing computational costs, but a trade‑off must be found between performance, coverage, and efficiency. GBC remains the most effective meth‑ od for rare attacks but, due to its computational cost, is poorly suited for embedded systems. ET and RF represent an excellent compromise between accuracy and speed. SC, while the most effective, requires significant resources. The scientific novelty of the research lies in the systematic evaluation of hierarchical and ensemble approaches for IDS in IoT, paving the way for creating more robust architectures adapted to IoT cybersecurity tasks.

Abstract The Qinghai-Tibet Plateau (QTP) is highly vulnerable to climate change, with permafrost changes exerting substantial impacts on regional ecosystems, hydrological stability, and cryospheric water resources. This study combines Coupled Model Intercomparison Project Phase 6 (CMIP6) data with four machine learning algorithms, including Support Vector Regression (SVR), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), and Deep Neural Network (DNN), to construct optimal models for predicting permafrost extent and maximum seasonal soil freeze depth (SFD) across the QTP. The relative importance and marginal effects of predictors were further evaluated using Random Forest (RF) analysis and Shapley Additive exPlanations (SHAP). Permafrost evolution from 2025 to 2100 was projected under four Shared Socioeconomic Pathway scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5). Results revealed that the DNN (R 2 = 0.961) outperforms other models in capturing frozen-ground dynamics. Key drivers of SFD were identified as summer-autumn precipitation, temperature, and non-frozen season soil moisture. Permafrost is projected to continuously degrade into seasonally frozen ground, with transition zones expanding from plateau margins toward the interior, leading to increased spatial fragmentation. SFD is expected to decline at rates strongly correlated with emission scenarios, with the most pronounced reductions along current permafrost margins and gradually extending inland. Notably, permafrost stability in high-altitude regions will face increasing challenges. The Northern Tibetan Plateau, the margins of the Qaidam Basin, and the foothills of the Gangdise Mountains were identified as high-risk zones for frozen-ground degradation under continued CO 2 emission trajectories.

Quantitative remote wound monitoring has the potential to shorten patient recovery time and alleviate the workload of healthcare professionals. In this study, a nitrogen-doped horizontally grown graphene (NHG) antenna sensor with a working frequency of 2.45 GHz was designed for wireless real-time monitoring of wounds. The sensor comprises 32 NHG microtubes (1 mm in diameter), a porous Cu radiation electrode, a polydimethylsiloxane substrate with a cylindrical channel array, and a Cu ground plane. Its novel structure enables body fluid and its temperature and pH value sensing by tracking dual signals, such as resonance frequency and return loss, thereby facilitating the identification of living organisms and real-time quantitative wound assessment. Notably, the NHG microtubes, which penetrate the Cu electrode and PDMS substrate, regulate the radiofrequency radiation field and enhance the monitoring sensitivity. The sensor exhibits a minimum fluid response volume of 25 μL, a temperature detection range of 34-43 °C, a resolution of 0.1 °C, and a response time of 20 s. Furthermore, the NHG antenna sensor reliably evaluated the pH value, volume, and area of the wound using a machine learning algorithm. The system was successfully validated for real-time monitoring of wound healing in mice and has been preliminarily applied to monitor wounds of various sizes and locations in human patients.

Quantum assisted machine learning (QML) have the potential to outperform the classical Machine Learning models by utilizing the quantum feature maps embeddings and variational circuits. In this study, a Hybrid Quantum Support Vector Machine (QSVM) and Quantum Neural Network (QNN) is proposed for supervised classification applied to the Breast Cancer Wisconsin Dataset. The QSVM component employs fidelity quantum kernels with three distinct feature maps—Z, ZZ and Pauli—to transform classical data into quantum states. The QNN is implemented using a variational quantum classifier based on the EstimatorQNN, which enables quantum-trainable parameters for enhanced feature extraction. The proposed hybrid methodology is evaluated using seven principal features, and experiments are conducted on IBM’s state vector Qiskit Aer local simulator. Performance is assessed based on accuracy, recall, precision, F1-score, and confusion matrix. Additionally, statistical significance is evaluated using paired t test and F test to compare the performance of different quantum feature maps and to validate improvements made in the Hybrid Model. The results demonstrate that the hybrid QSVM/QNN approach improves classification performance not only in terms of accuracy by achieving above 90% accuracy but also by reducing computational cost. This work contributes to the advancement of quantum enhanced machine learning (ML) algorithms by demonstrating the hybrid quantum-classical approach in structured data classification.