Search Results (937)

Resource-limited settings continue to face challenges in the identification of COVID-19, bacterial pneumonia, viral pneumonia, and normal lung conditions because of the overlap of CT appearance and inter-observer variability. We justify a hybrid architecture of deep learning which combines hand-designed descriptors (Histogram of Oriented Gradients, Local Binary Patterns) and a 20-layer Convolutional Neural Network with dual self-attention. Handcrafted features were then trained with Support Vector Machines, and ensemble averaging was used to integrate the results with the CNN. The confidence level of 0.7 was used to mark suspicious cases to be reviewed manually. On a balanced dataset of 14,000 chest CT scans (3500 per class), the model was trained and cross-validated five-fold on a patient-wise basis. It had 97.43% test accuracy and a macro F1-score of 0.97, which was statistically significant compared to standalone CNN (92.0%), ResNet-50 (90.0%), multiscale CNN (94.5%), and ensemble CNN (96.0%). A further 2–3% enhancement was added by the self-attention module that targets the diagnostically salient lung regions. The predictions that were below the confidence limit amounted to only 5 percent, which indicated reliability and clinical usefulness. The framework provides an interpretable and scalable method of diagnosing multiclass lung disease, especially applicable to be deployed in healthcare settings with limited resources. The further development of the work will involve the multi-center validation, optimization of the model, and greater interpretability to be used in the real world. Full article

Background/Objectives: Obstructive sleep apnea (OSA) is a prevalent, yet underdiagnosed, disorder associated with cardiovascular and cognitive risks. While overnight polysomnography (PSG) remains the diagnostic gold standard, it is resource-intensive and impractical for large-scale rapid screening. Methods: This study extends prior work on feature extraction and binary classification using tracheal breathing sounds (TBS) and anthropometric data by introducing a meta-modeling framework that utilizes machine learning (ML) and aggregates six one-vs.-one classifiers for multi-class OSA severity prediction. We employed out-of-bag (OOB) estimation and three-fold cross-validation to assess model generalization performance. To enhance reliability, the framework incorporates conformal prediction to provide calibrated confidence sets. Results: In the three-class setting (non, mild, moderate/severe), the model achieved 76.7% test accuracy, 77.7% sensitivity, and 87.1% specificity, with strong OOB performance of 91.1% accuracy, 91.6% sensitivity, and 95.3% specificity. Three-fold confirmed stable performance across folds (mean accuracy: 77.8%; mean sensitivity: 78.6%; mean specificity: 76.4%) and conformal prediction achieved full coverage with an average set size of 2. In the four-class setting (non, mild, moderate, severe), the model achieved 76.7% test accuracy, 75% sensitivity, and 92% specificity, with OOB performance of 88.2% accuracy, 91.6% sensitivity, and 88.2% specificity. Conclusions: These findings support the potential of this non-invasive system as an efficient and rapid OSA severity assessment whilst awake, offering a scalable alternative to PSG for large-scale screening and clinical triaging. Full article

This study proposes a multi-objective, multi-class explainable modeling framework to explain country performance profiles in PISA Mathematics (PISAM), Reading (PISAR), and Science (PISAS). Instead of treating PISA as a simple ranking, the study models each country’s Low/Medium/High-achieving class and asks which structural signals the model relies on when assigning a country to this class. To this end, the study combines governance quality (e.g., accountability, control of corruption, and political stability, etc.), economic and administrative capacity, and regional/institutional location in a single prediction pipeline and explains the resulting classifications with SHAP contributions conditional on class. While the findings do not point to a single, universal determinant, in mathematics, high-level profiles cluster around political stability, economic scale barriers, and regional location, along with governance indicators; in reading, economic capacity is explicitly integrated into this institutional core; and in science, in addition to these two dimensions, the shared institutional dynamics of regional blocs come into play. Furthermore, the study not only produces explanations but also quantitatively reports their reliability. The fit with the model output (Fidelity) and the traceability of the decision logic (Faithfulness) are 0.95/0.85 for PISAM, 0.89/0.92 for PISAR, and 0.89/0.89 for PISAS, which demonstrates high internal consistency and traceability of the decision process. Overall, the study reframes the PISA results not as isolated test scores but as structural profiles generated by the combination of governance, capacity, and region, revealing the policy-relevant levers behind “high performance” as a transparent and reproducible decision-making pipeline. This provides policymakers with an important roadmap for creating a sustainable education policy. Full article

Emerging from millions of years of evolutionary optimization, Natural products (NPs) remain unique, unparalleled sources of bioactive scaffolds. Unlike synthetic molecules engineered around single therapeutic targets, NPs often exhibit multi-target, system-level bioactivity, aligned with the principles of network pharmacology, which modulates pathways in a coordinated, non-disruptive manner. This approach reduces resistance, buffers compensatory feedback loops, and enhances therapeutic resilience. Fungal endophytes represent one of the most chemically diverse and biologically sophisticated NP reservoirs known, producing polyketides, alkaloids, terpenoids, and peptides with intricate three-dimensional architectures and emergent bioactivity patterns that remain exceptionally difficult to design de novo. Advances in artificial intelligence (AI), machine learning, deep learning, and multi-omics integration have redefined the discovery landscape, transforming previously intractable fungal metabolomes and cryptic biosynthetic gene clusters (BGCs) into tractable, predictable, and engineerable systems. AI accelerates genome mining, metabolomic annotation, BGC-metabolite linking, structure prediction, and activation of silent pathways. Generative AI and diffusion models now enable de novo design of NP-inspired scaffolds while preserving biosynthetic feasibility, opening new opportunities for direct evolution, pathway refactoring, and precision biomanufacturing. This review synthesizes the chemical and biosynthetic diversity of major NP classes from fungal endophytes and maps them onto the rapidly expanding ecosystem of AI-driven tools. We outline how AI transforms NP discovery from empirical screening into a predictive, hypothesis-driven discipline with direct industrial implications for drug discovery and synthetic biology. By coupling evolutionarily refined chemistry with modern computational intelligence, the field is poised for a new era in which natural-product leads are not only rediscovered but systematically expanded, engineered, and industrialized to address urgent biomedical and sustainability challenges. Full article

For mine ventilation-on-demand (VOD) scenarios, conventional joint optimisation of airflow augmentation and energy saving in mine ventilation systems is often constrained in practical engineering applications by shrinkage of the feasible region, limited adjustable resistance margins, and strongly multi-modal objective functions. These factors tend to result in low solution efficiency, pronounced sensitivity to initial values and insufficient solution robustness. In response to these challenges, a two-layer intelligent evolutionary optimisation framework, termed ES–Hybrid JADE with Competitive Niching, is developed in this study. In the outer layer, four classes of evolutionary algorithms—CMAES, DE, ES, and GA—are comparatively assessed over 50 repeated test runs, with a combined ranking based on convergence speed and solution quality adopted as the evaluation metric. ES, with a rank_mean of 2.0, is ultimately selected as the global hyper-parameter self-adaptive regulator. In the inner layer, four algorithms—COBYLA, JADE, PSO and TPE—are compared. The results indicate that JADE achieves the best overall performance in terms of terminal objective value, multi-dimensional performance trade-offs and robustness across random seeds. Furthermore, all four inner-layer algorithms attain feasible solutions with a success rate of 1.0 under the prescribed constraints, thereby ensuring that the entire optimisation process remains within the feasible domain. The proposed framework is applied to an exhaust-type dual-fan ventilation system in a coal mine in Shaanxi Province as an engineering case study. By integrating GA-based automatic ventilation network drawing (longest-path/connected-path) with roadway sensitivity analysis and maximum resistance increment assessment, two solution schemes—direct optimisation and composite optimisation—are constructed and compared. The results show that, within the airflow augmentation interval [0.40, 0.55], the two schemes are essentially equivalent in terms of the optimal augmentation effect, whereas the computation time of the composite optimisation scheme is reduced significantly from approximately 29 min to about 13 s, and a set of multi-modal elite solutions can be provided to support dispatch and decision-making. Under global constraints, a maximum achievable airflow increment of approximately 0.66 m³·s⁻¹ is obtained for branch 10, and optimal dual-branch and triple-branch cooperative augmentation combinations, together with the corresponding power projections, are further derived. To the best of our knowledge, prior VOD airflow-augmentation studies have not combined feasibility-region contraction (via sensitivity- and resistance-margin gating) with a two-layer ES-tuned JADE optimiser equipped with Competitive Niching to output multiple feasible optima. This work provides new insight that the constrained airflow-augmentation problem is intrinsically multimodal, and that retaining multiple basins of attraction yields dispatch-ready elite solutions while achieving orders-of-magnitude runtime reduction through prediction-based constraints. The study demonstrates that the proposed two-layer intelligent evolutionary framework combines fast convergence with high solution stability under strict feasibility constraints, and can be employed as an engineering algorithmic core for energy-efficiency co-ordination in mine VOD control. Full article

This research entails comparative analysis and optimisation of machine learning models for regression and classification tasks on structured tabular datasets. The primary target audience for this analysis comprises researchers and practitioners working with structured tabular data. Common fields include biostatistics, insurance, and financial risk modelling, where computational efficiency and robust predictive performance are essential. Four machine learning techniques (i.e., linear/logistic regression, support vector machines (SVMs), Extreme Gradient Boosting (XGBoost), and Multi-Layered Perceptrons (MLPs)) were applied across 72 datasets sourced from OpenML and Kaggle. The datasets systematically varied by observation size, dimensionality, noise levels, linearity, and class balance. Based on extensive empirical analysis (72 datasets $\times 4$ models $\times 2$ configurations $=576$ experiments), it is observed that, understanding the dataset characteristics is more critical than extensive hyperparameter tuning for optimal model performance. Also, linear models are robust across various settings, while non-linear models, like XGBoost and MLP, perform better in complex and noisy environments. In general, this study provides valuable insights for model selection and benchmarking in machine learning applications that involve structured tabular datasets. Full article

This study presents an integrated framework that couples three-dimensional geotechnical ground modeling with a HAZUS-based urban seismic vulnerability assessment for Seoul, Korea. Over 63,000 boreholes, in situ seismic tests, and building inventory records were compiled into a unified relational database following rigorous multi-stage quality control. A multi-parameter N–Vs regression model was calibrated to supplement missing shear-wave velocity (Vs) data, reducing prediction errors by more than 20% relative to conventional empirical equations. Based on the quality-controlled Vs dataset, a high-resolution three-dimensional Vs–ground model was constructed to represent subsurface heterogeneity and associated uncertainty across the metropolitan area. The building inventory, comprising approximately 700,000 structures, was standardized according to the HAZUS structural taxonomy and mapped to Korean seismic design eras, enabling a Seoul-adapted vulnerability assessment in which exposure characterization and seismic demand are localized. Site-specific ground-motion amplification and response spectra derived from the 3D ground model were used to modify the spectral acceleration input to the HAZUS fragility functions. Results reveal pronounced spatial variability in site conditions, with northern mountainous zones corresponding primarily to NEHRP Site Class B, central districts to Class C, and southern alluvial basins to Classes D–E, producing amplification differences of up to 1.7 under identical input spectral accelerations. High-risk zones such as Gangnam, Songpa, and Yeouido exhibit concentrated expected damage where thick alluvial deposits coincide with dense stocks of mid-rise reinforced-concrete buildings. Overall, the study demonstrates that integrating high-resolution 3D geotechnical ground models with HAZUS-based fragility analysis provides a physically consistent and data-driven basis for urban-scale seismic risk assessment and resilience planning. Full article

Background: Accurate staging of bladder cancer is critical for guiding clinical management, particularly the distinction between non–muscle-invasive (T1) and muscle-invasive (T2–T4) disease. Although MRI offers superior soft-tissue contrast, image interpretation remains opera-tor-dependent and subject to inter-observer variability. This study proposes an automated deep learning framework for MRI-based bladder cancer staging to support standardized radio-logical interpretation. Methods: A sequential AI-based pipeline was developed, integrating hybrid tumor segmentation using YOLOv11 for lesion detection and DeepLabV3 for boundary refinement, followed by three deep learning classifiers (VGG19, ResNet50, and Vision Transformer) for MRI-based stage prediction. A total of 416 T2-weighted MRI images with radiology-derived stage labels (T1–T4) were included, with data augmentation applied during training. Model performance was evaluated using accuracy, precision, recall, F1-score, and multi-class AUC. Performance un-certainty was characterized using patient-level bootstrap confidence intervals under a fixed training and evaluation pipeline. Results: All evaluated models demonstrated high and broadly comparable discriminative performance for MRI-based bladder cancer staging within the present dataset, with high point estimates of accuracy and AUC, particularly for differentiating non–muscle-invasive from muscle-invasive disease. Calibration analysis characterized the probabilistic behavior of predicted stage probabilities under the current experimental setting. Conclusions: The proposed framework demonstrates the feasibility of automated MRI-based bladder cancer staging derived from radiological reference labels and supports the potential of deep learning for stand-ardizing and reproducing MRI-based staging procedures. Rather than serving as an independent clinical decision-support system, the framework is intended as a methodological and work-flow-oriented tool for automated staging consistency. Further validation using multi-center datasets, patient-level data splitting prior to augmentation, pathology-confirmed reference stand-ards, and explainable AI techniques is required to establish generalizability and clinical relevance. Full article

Background: Accurate and dependable brain tumor classification from magnetic resonance imaging (MRI) is essential for clinical decision support, yet remains challenging due to inter-dataset variability, heterogeneous tumor appearances, and limited generalization of many deep learning models. Existing studies often rely on single-dataset evaluation, insufficient statistical validation, or lack interpretability, which restricts their clinical reliability and real-world deployment. Methods: This study proposes a robust brain tumor classification framework based on the ConvNeXt Base architecture. The model is evaluated across three independent MRI datasets comprising four classes—glioma, meningioma, pituitary tumor, and no tumor. Performance is assessed using class-wise and aggregate metrics, including accuracy, precision, recall, F1-score, AUC, and Cohen’s Kappa. The experimental analysis is complemented by ablation studies, computational efficiency evaluation, and rigorous statistical validation using Friedman’s aligned ranks test, Holm and Wilcoxon post hoc tests, Kendall’s W, critical difference diagrams, and TOPSIS-based multi-criteria ranking. Model interpretability is examined using Grad-CAM++ and Gradient SHAP. Results: ConvNeXt Base consistently achieves near-perfect classification performance across all datasets, with accuracies exceeding 99.6% and AUC values approaching 1.0, while maintaining balanced class-wise behavior. Statistical analyses confirm that the observed performance gains over competing architectures are significant and reproducible. Efficiency results demonstrate favorable inference speed and resource usage, and explainability analyses show that predictions are driven by tumor-relevant regions. Conclusions: The results demonstrate that ConvNeXt Base provides a reliable, generalizable, and explainable solution for MRI-based brain tumor classification. Its strong diagnostic accuracy, statistical robustness, and computational efficiency support its suitability for integration into real-world clinical and diagnostic workflows. Full article

Salmonellosis remains a persistent threat to global food safety and poultry productivity, compounded by rising antimicrobial resistance. Here, we report the in silico design and immunoinformatic validation of a broad-spectrum, edible multi-epitope vaccine targeting conserved adhesion and biofilm-associated proteins (FimH, AgfA, SefA, SefD, and MrkD) of Salmonella spp. Two constructs were engineered by integrating cytotoxic (CTL) and helper (HTL) epitopes with β-defensin-3 (HBD-3) or lipopolysaccharide (LPS) adjuvants, optimized for expression in Chlorella vulgaris. Structural modeling confirmed native-like folding (z-scores −2.58 and −5.22) and high stability indices. Molecular docking and dynamics revealed that the LPS-adjuvanted construct (Construct 2) forms a highly stable complex with Toll-like receptor 3 (HADDOCK score −63.4; desolvation energy −50.2 kcal/mol), indicating potent innate immune activation. Immune simulations predicted strong IgM-to-IgG class switching and durable humoral responses, consistent with effective antigen clearance. Codon optimization achieved high adaptability for algal expression (CAI = 0.93; GC ≈ 65%), supporting scalable microalgae-based production. Compared with current parenteral vaccines, offering a low-cost, non-invasive way to curb Salmonella in poultry, this edible vaccine platform reduces dependence on antibiotics. Our approach, which combines computational vaccinology with a safe-by-design sustainable biomanufacturing perspective, outlines a One Health framework for advancing antimicrobial stewardship and food safety. Full article

Refill Friction Stir Spot Welding (RFSSW) provides significant advantages over competing spot joining technologies, but detecting RFSSW’s often small and subtle defects remains challenging. In this study, kinematic feedback data from a RFSSW machine’s factory-installed sensors was used to successfully predict defect presence with 96% accuracy (F1 = 0.92) and preliminary multi-class defect diagnosis with 84% accuracy (F1 = 0.82). Thirty adverse treatments (e.g., contaminated coupons, worn tools, and incorrect material thickness) were carried out to create 300 potentially defective welds, plus control welds, which were then evaluated using profilometry, computed tomography (CT) scanning, cutting and polishing, and tensile testing. Various machine learning (ML) models were trained and compared on statistical features, with support vector machine (SVM) achieving top performance on final quality prediction (binary), random forest outperforming other models in classifying welds into six diagnosis categories (plus a control category) based on the adverse treatments. Key predictors linking process signals to defect formation were identified, such as minimum spindle torque during the plunge phase. In conclusion a framework is proposed to integrate these models into a manufacturing setting for low-cost, full-coverage evaluation of RFSSWs. Full article

The rapid advancement of sequencing technologies has led to the generation of complex multi-omics data, which are often high-dimensional, noisy, and imbalanced, posing significant challenges for traditional machine learning methods. The novelty of this work resides in the architecture-level integration of autoencoders with Generative Adversarial Network (GAN) and Conditional Tabular Generative Adversarial Network (CTGAN) models, where the autoencoder is employed for latent feature extraction and noise reduction, while GAN-based models are used for realistic sample generation and class imbalance mitigation in multi-omics cancer datasets. This study proposes a novel framework that combines an autoencoder for dimensionality reduction and a CTGAN for generating synthetic samples to balance underrepresented classes. The process starts with selecting the most discriminative features, then extracting latent representations for each omic type, merging them, and generating new minority samples. Finally, all samples are used to train a neural network to predict specific cancer outcomes, defined here as clinically relevant biomarkers or patient characteristics. In this work, the considered outcome in the bladder cancer is Tumor Mutational Burden (TMB), while the breast cancer outcome is menopausal status, a key factor in treatment planning. Experimental results show that the proposed model achieves high precision, with an average precision of 0.9929 for TMB prediction in bladder cancer and 0.9748 for menopausal status in breast cancer, and reaches perfect precision (1.000) for the positive class in both cases. In addition, the proposed AE–CTGAN framework consistently outperformed an autoencoder combined with a standard GAN across all evaluation metrics, achieving average accuracies of 0.9929 and 0.9748, recall values of 0.9846 and 0.9777, and F1-scores of 0.9922 for bladder and breast cancer datasets, respectively. A comparative fidelity analysis in the latent space further demonstrated the superiority of CTGAN, reducing the average Euclidean distance between real and synthetic samples by approximately 72% for bladder cancer and by up to 84% for breast cancer compared to a standard GAN. These findings confirm that CTGAN generates high-fidelity synthetic samples that preserve the structural characteristics of real multi-omics data, leading to more reliable class balancing and improved predictive performance. Overall, the proposed framework provides an effective and robust solution for handling class imbalance in multi-omics cancer data and enhances the accuracy of clinically relevant outcome prediction. Full article

We aimed to establish a rapid and precise method for identifying and quantifying multi-species vegetable oil (corn oil, olive oil (OLO), soybean oil, and sunflower oil (SUO)) adulterations in camellia oil (CAO), using soft ionization by chemical reaction in transfer–high-resolution mass spectrometry (SICRIT-HRMS) and machine learning methods. The results showed that SICRIT-HRMS could effectively characterize the volatile profiles of pure and adulterated CAO samples, including binary, ternary, quaternary, and quinary adulteration systems. The low m/z region (especially 100–300) exhibited importance to oil classification in multiple feature-selection methods. For qualitative detection, binary classification models based on convolutional neural networks (CNN), Random Forest (RF), and gradient boosting trees (GBT) algorithms showed high accuracies (98.70–100.00%) for identifying CAO adulteration under no dimensionality reduction (NON), principal component analysis (PCA), and uniform manifold approximation and projection (UMAP) strategies. The RF algorithm exhibited relatively high accuracy (96.25–99.45%) in multiclass classification. Moreover, the five models, including CNN, RF, support vector machines (SVM), logistic regression (LR), and GBT, exhibited different performances in distinguishing pure and adulterated CAO. Among 1093 blind oil samples, under NON, PCA, and UMAP: 10, 5, and 67 samples were misclassified by CNN model; 6, 7, and 41 samples were misclassified by RF model; 8, 9, and 82 samples were misclassified by SVM model; 17, 18, and 78 samples were misclassified by LR model; 7, 9, and 43 samples were misclassified by GBT model. For quantitative prediction, the PCA-CNN model performed optimally in predicting adulteration levels in CAO, especially with respect to OLO and SUO, exhibiting a high coefficient of determination for calibration ($R_C^2$, 0.9664–0.9974) and coefficient of determination for prediction ($R_p^2$, 0.9599–0.9963) values, low root mean square error of calibration (RMSEC, 0.9–5.3%) and root mean square error of prediction (RMSEP, 1.1–5.8%) values, and RPD (5.0–16.3) values greater than 3.0. These results indicate that SICRIT-HRMS combined with machine learning can rapidly and accurately identify and quantify multi-species vegetable oil adulterations in CAO, which provides a reference for developing non-targeted and high-throughput detection methods in edible oil authenticity. Full article

Accurately identifying drug off-targets is essential for reducing toxicity and improving the success rate of pharmaceutical discovery pipelines. However, current deep learning approaches often struggle to fuse chemical structure, protein biology, and multi-target context. Here, we introduce HDPC-LGT (Hybrid Dual-Prompt Cross-Attention Ligand–Protein Graph Transformer), a framework designed to predict ligand binding across sixteen human translation-related proteins clinically associated with antibiotic toxicity. HDPC-LGT combines graph-based chemical reasoning with protein language model embeddings and structural priors to capture biologically meaningful ligand–protein interactions. The model was trained on 216,482 experimentally validated ligand–protein pairs from the Chemical Database of Bioactive Molecules (ChEMBL) and the Protein–Ligand Binding Database (BindingDB) and evaluated using scaffold-level, protein-level, and combined holdout strategies. HDPC-LGT achieves a macro receiver operating characteristic–area under the curve (macro ROC–AUC) of 0.996 and a micro F1-score (micro F1) of 0.989, outperforming Deep Drug–Target Affinity Model (DeepDTA), Graph-based Drug–Target Affinity Model (GraphDTA), Molecule–Protein Interaction Transformer (MolTrans), Cross-Attention Transformer for Drug–Target Interaction (CAT–DTI), and Heterogeneous Graph Transformer for Drug–Target Affinity (HGT–DTA) by 3–7%. External validation using the Papyrus universal bioactivity resource (Papyrus), the Protein Data Bank binding subset (PDBbind), and the benchmark Yamanishi dataset confirms strong generalisation to unseen chemotypes and proteins. HDPC-LGT also provides biologically interpretable outputs: cross-attention maps, Integrated Gradients (IG), and Gradient-weighted Class Activation Mapping (Grad-CAM) highlight catalytic residues in aminoacyl-tRNA synthetases (aaRSs), ribosomal tunnel regions, and pharmacophoric interaction patterns, aligning with known biochemical mechanisms. By integrating multimodal biochemical information with deep learning, HDPC-LGT offers a practical tool for off-target toxicity prediction, structure-based lead optimisation, and polypharmacology research, with potential applications in antibiotic development, safety profiling, and rational compound redesign. Full article

This study develops a comprehensive workflow integrating Headspace Solid-Phase Microextraction Gas Chromatography–Mass Spectrometry (HS-SPME-GC-MS) with advanced supervised machine learning to authenticate the botanical origin of honeys from five distinct floral sources—coriander, orange blossom, astragalus, rosemary, and chehelgiah. While HS-SPME-GC-MS combined with traditional chemometrics (e.g., PCA, LDA, OPLS-DA) is well-established for honey discrimination, the application and direct comparison of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Neural Network (NN) models represent a significant advancement in multiclass prediction accuracy and model robustness. A total of 57 honey samples were analyzed to generate detailed volatile organic compound (VOC) profiles. Key chemotaxonomic markers were identified: anethole in coriander and chehelgiah, thymoquinone in astragalus, p-menth-8-en-1-ol in orange blossom, and dill ester (3,6-dimethyl-2,3,3a,4,5,7a-hexahydrobenzofuran) in rosemary. Principal component analysis (PCA) revealed clear separation across botanical classes (PC1: 49.8%; PC2: 22.6%). Three classification models—RF, XGBoost, and NN—were trained on standardized, stratified data. The NN model achieved the highest accuracy (90.32%), followed by XGBoost (86.69%) and RF (83.47%), with superior per-class F1-scores and near-perfect specificity (>0.95). Confusion matrices confirmed minimal misclassification, particularly in the NN model. This work establishes HS-SPME-GC-MS coupled with deep learning as a rapid, sensitive, and reliable tool for multiclass honey botanical authentication, offering strong potential for real-time quality control, fraud detection, and premium market certification. Full article