The rapid evolution of Domain Generation Algorithm (DGA)-driven attacks and obfuscated DNS traffic exposes fundamental weaknesses in conventional machine learning-based threat detection systems, particularly under adversarial manipulation. This study introduces FGM-GAN, a hybrid adversarial learning framework that synergistically combines gradient-based Fast Gradient Method (FGM) perturbations with adaptive Generative Adversarial Network (GAN)-based perturbations to improve both robustness and interpretability of deep neural networks for DNS threat classification. Unlike existing adversarial defenses that rely on model-specific perturbations, FGM-GAN explicitly learns class-conditional adversarial distributions for benign, phishing, and malware domains. This design enables the generation of realistic, feature-aligned perturbations that exhibit strong cross-model transferability. Experiments were conducted on the 32-feature CIC-BELL-DNS-2021 dataset (approximately 7000 labeled samples) using 5-fold cross-validation, hybrid perturbations with and , and evaluated against baseline DNN, SVM, Random Forest, KNN, and Decision Tree classifiers using accuracy and robustness metrics. Comprehensive evaluation demonstrates that FGM-GAN consistently improves robustness across diverse adversarial attacks (FGM, PGD, MIM, C&W) while maintaining stable performance across folds. Ablation studies and reduced-capacity variants confirm that gains arise from the hybrid adversarial mechanism rather than over-parameterization or hyperparameter tuning, and statistical significance tests verify the reproducibility of results. To enhance transparency and operational trust, the framework integrates multi-level explainable AI analyses spanning feature, neuron, and layer representations. These analyses consistently identify a compact set of high-impact DNS features and reveal structured adversarial propagation patterns, showing that robustness emerges from semantically meaningful representation learning. Collectively, these findings position FGM-GAN as a scalable and interpretable adversarial learning solution that jointly addresses robustness, transferability, and explainability in real-world DNS-based cybersecurity environments. • FGM-GAN hybrid improves neural network robustness against adversarial attacks • GANs produce realistic, class-specific adversarial perturbations for DNS data • Adversarial transferability validated across KNN, SVM, Decision Trees, RF • Gradient-XAI interprets feature, neuron, and layer-level model vulnerabilities • Combines robustness and explainability for actionable cyber threat intelligence

The aim of the study was to develop a model using machine learning algorithms for diagnosing complete and incomplete injury from heart rate variation (HRV) parameters and other easily obtained demographics. Random Forest, Decision Tree, KNN Classifier, Gradient Boost, Bagging Classifier, Support Vector Machine, Voting Classifier, XGB Classifier, MLP Classifier and feedforward neural networks were trained on 296 sets of patient data including 11 HRV parameters. Feature selection methods were used to improve models and identify parameters with high contribution to model predictions. The feature selected MLPClassifier achieved an accuracy of 85.33%, with an AUC of 0.8590 while the neural network model had an accuracy of 86.67% with an AUC of 0.9608. Location of spinal injury, age, mean heart rate and mean R-R interval were the greatest contributors to the machine learning models. This study demonstrated that machine learning algorithms trained on HRV data could be an invaluable tool for diagnosing and monitoring people with SCI and overall improving their quality of life.

Delayed switching faults in hydraulic directional control valves can significantly degrade system performance and reliability, yet their diagnosis remains challenging due to complex fault mechanisms and coupled sensor responses and limited fault samples in industrial applications. While data-driven approaches, including deep learning-based methods, have shown promising performance in fault diagnosis, their practical deployment in industrial quality inspection and condition monitoring is often constrained by limited fault data availability and insufficient physical interpretability of the diagnostic results. In this study, an interpretable fault diagnosis framework for delayed switching faults in hydraulic directional control valves is proposed based on a simulation-guided feature construction method and multi-pressure signal analysis. Instead of using simulation to generate synthetic training data, a physical simulation model is employed to analyze fault mechanisms and to guide the design of valve-level diagnostic features derived from inter-sensor pressure differences. These features are further evaluated using several classical machine learning classifiers, including RF, SVM, KNN, and LR under conditions of limited fault samples. Experimental results demonstrate that the proposed method effectively captures the structural imbalance caused by internal valve faults and achieves high diagnostic accuracy and robustness compared with conventional single-sensor approaches and purely data-driven black-box models. The proposed framework provides a practical and physically interpretable solution for hydraulic valve fault diagnosis under small-sample conditions and offers potential value for industrial quality inspection and maintenance applications.

This study proposes a multi-layered framework to quantify gender and ethnic bias in state-of-the-art text-to-image models using geographically grounded evaluations across ten American countries. We evaluate Stable Diffusion 1.5, SDXL, Gemini, Flux, and DALL.E by creating two datasets: one organized by race–gender categories and another using only nationality prompts. Using CLIP embeddings, we extract demographic representations through three methods: (i) a KNN classifier trained solely on model-generated demographic exemplars; (ii) a distance-based comparison of country-level embedding centroids to race–gender centroids; and (iii) a semantic layer from a multimodal LLM that infers demographic attributes from each portrait. An autoencoder provides latent-space visualizations for interpretable geometric inspection of cross-country patterns. Across models, we find systematic deviations from real census distributions, including consistent male overrepresentation (except Gemini, which skews female) and strong ethnic homogenization. Latin American countries are repeatedly mapped to Indigenous archetypes, while Canada and the United States are largely rendered with White prototypes. These distortions appear in both embedding geometry and LLM-based semantic judgments, indicating that bias affects low-level visual structure and high-level interpretation. By linking synthetic portraits to demographic baselines, this study offers a scalable method to audit representational fairness in AI models and stresses the need for culturally contextualized evaluation in responsible AI development.

ABSTRACT Neural recording electrodes are pivotal components of a brain‐computer interface (BCI) for diagnosing neurological diseases. However, the conventional electroencephalography (EEG) electrodes face challenges in long‐term stability and biocompatibility. The contemporary dry electrodes offer enhanced stability but remain suboptimal in recording sensitivity. To address these constraints, we developed mechanically flexible, high‐sensitivity, biocompatible electrodes employing a redox‐active multienzyme‐mimetic gold cluster integrated into highly conductive polymer hydrogel networks for prolonged neural monitoring in Parkinson's disease (PD). The resulting cluster‐hydrogel (CH) electrode exhibits an ultralow contact impedance of 6.9 kΩ·cm 2 , equivalent to 1/6 th that of conventional Ag/AgCl electrodes and 1/24 th that of metallic electrodes, while sustaining a high signal‐to‐noise ratio of 6.6 dB over 12 h continuous recording. Clinically, through the integration of machine learning techniques, the CH electrodes enabled accurate PD diagnosis via alpha reactivity and entropy analyses, achieving an area under the curve (AUC) of 0.9 (KNN classifier) and a phase‐locking value (PLV) Pearson correlation of 0.934. In addition, the CH electrodes outperformed conventional EEG electrodes in alpha‐band sensitivity, achieving a 16.2‐fold higher signal‐to‐noise ratio that enhanced the detection of clinically diagnostic electrophysiological signatures in PD. Combined with high biocompatibility, the CH electrodes hold translational promise as an efficient, safe, and stable neural recording system for clinical BCIs and neuroscience applications.

The Internet of Things (IoT) has emerged as a pervasive technological paradigm that interconnects heterogeneous devices and sensors, enabling continuous data acquisition, communication, and intelligent decision-making. However, the large-scale, dynamic, and heterogeneous nature of IoT environments introduces significant cybersecurity threats, making intrusion detection a critical component of IoT network protection. The complexity and high dimensionality of IoT traffic data pose substantial challenges for machine-learning-based intrusion detection systems, particularly for classification accuracy. In this context, feature selection (FS), which aims to identify the most informative and non-redundant features, plays a vital role in enhancing detection performance. This study proposes a model to investigate the FS problem using bird-based metaheuristic optimization algorithms, integrated with a taper-shaped transfer function for binary transformation. The proposed framework aims to identify the most informative and non-redundant features from high-dimensional IoT datasets to enhance classification performance. The kNN, SVM, and RF classifiers are employed to evaluate the model using 10-fold cross-validation. Experimental results on the RT-IoT2022 and IoTID20 datasets show that bird-based FS methods achieve substantial dimensionality reduction and strong classification performance. The Secretary Bird Optimization Algorithm (SBOA), the best-performing model on the RT-IoT2022, identified only 6 of 81 features, achieving the highest feature reduction ratio of 92.59% and a classification accuracy of 99.69%. Moreover, the algorithm selected only 7 of 81 features, achieving a feature reduction ratio of 91.36% and a classification accuracy of 98.46% on the IoTID20 dataset. Additionally, SBOA performs well in terms of sensitivity, specificity, precision, and computational time, underscoring its robustness in handling complex IoT traffic data. The findings indicate that bird-inspired optimization approaches, when integrated with an effective binary transfer mechanism, offer a powerful solution for real-time IoT intrusion detection systems.

Background/Objectives: Brain tumors are among the most severe neurological disorders, and their variability in size, morphology, and anatomical location complicates early and accurate diagnosis. Although magnetic resonance imaging (MRI) is the most reliable non-invasive modality for tumor detection, manual interpretation remains time-consuming, subjective, and susceptible to human error. This study aims to develop an optimization-driven hybrid machine learning framework for accurate and computationally efficient automatic brain tumor classification. Methods: The dataset includes 834 MRI images (583-training, 123-validation, 128-independent test). Because YOLOv11 detects tumor and non-tumor regions separately, the sample size doubled during region-based analysis, and all subsequent stages were conducted at the regions of interest (ROI) level. On the independent test set, YOLOv11 achieved 98.87% mAP@50, 98.54% precision, and 98.21% recall. The proposed framework combines automated tumor localization with image standardization using Gaussian noise reduction and bilinear interpolation. From the processed MR images, 39 entropy-based features were extracted. To enhance diagnostic performance and eliminate redundant information, the superb fairy-wren optimization algorithm (SFOA) was applied for feature selection and compared with particle swarm optimization (PSO), Harris hawk optimization (HHO), and puma optimization (PO). Final classification was primarily performed using k-nearest neighbors (kNN), while support vector machines (SVM) were used for comparative evaluation. Results: SFOA reduced the feature dimensionality from 39 to 5 features while achieving 99.20% classification accuracy on the independent test set. In comparison, PSO selected 10 features, HHO selected 6 features and PO selected 10 features, all achieving 98.45% accuracy. The best performance obtained with SVM was 98.45% accuracy (HHO-SVM), which remained lower than the 99.20% achieved by the proposed SFOA-kNN model. Conclusions: The results indicate that combining entropy-based feature extraction with SFOA-driven feature selection and kNN classification significantly enhances diagnostic accuracy while reducing computational complexity, highlighting the strong potential of the proposed framework for integration into computer-aided diagnosis systems to support clinical decision-making.

ABSTRACT Rolling element bearing faults are a major source of failure in rotating machinery, making accurate and reliable fault diagnosis essential for condition monitoring and predictive maintenance. This study presents a supervised machine learning (ML) based framework for bearing fault classification using statistical features extracted from vibration signals. Standard preprocessing techniques, including feature extraction, feature standardisation, and label encoding, were employed to prepare the data for model training. Multiple classifiers were evaluated, and their performance was assessed using standard evaluation metrics and confusion matrix analysis. The results show that the Decision Tree (DT) classifier achieved the highest classification accuracy of 93.57%, indicating its strong capability to model nonlinear relationships between vibration features and bearing fault categories. The k-Nearest Neighbour (kNN) classifier recorded a slightly lower accuracy of 92.52% but achieved a high AUC of 98.60%, reflecting excellent class separability and generalisation capability. The strong AUC value suggests that kNN effectively distinguishes between healthy and faulty bearing conditions as well as among different fault categories when vibration features are properly scaled. Overall, the results indicate that while tree-based models offer higher raw classification accuracy, probabilistic and distance-based approaches provide strong generalisation performance and robust fault separability. The proposed approach shows strong generalisation capability, with only limited misclassification observed between certain fault types exhibiting similar vibration characteristics. Feature importance analysis further highlights the critical role of higher-order statistical features in distinguishing bearing health conditions. Overall, the findings confirm the effectiveness of the proposed framework for vibration-based bearing fault diagnosis and its potential applicability in real-world condition monitoring systems.

Retraction Note: Classification of yoga, meditation, combined yoga–meditation EEG signals using L-SVM, KNN, and MLP classifiers

Hybrid fractional thermoelastic–machine learning (KNN, CNN and SVM classifier) framework for heat and mass transfer: a computational mechanics approach

Hybrid fractional thermoelastic–machine learning (KNN, CNN and SVM classifier) framework for heat and mass transfer: A computational mechanics approach

The increasing interconnectivity of healthcare devices through the Internet of Medical Things (IoMT) has improved patient monitoring and treatment, but also exposed these systems to malicious cyberattacks that threaten both patient safety and data integrity. Existing machine learning (ML)-based approaches have attempted to detect such attacks, but most rely on all dataset features, including irrelevant and redundant ones, which increases computational cost and reduces detection accuracy. Feature selection techniques such as Particle Swarm Optimization (PSO) have been used to address this challenge, yet their default fitness functions fail to select the most suitable features for each classifier, often leading to suboptimal results. To overcome these limitations, this study introduces a novel fitness function integrated with PSO and ML classifiers to identify the most relevant features for accurate attack detection in IoMT devices. The proposed framework was evaluated using the NSL-KDD dataset (41 features) with RF, KNN, SVM, and LR classifiers. The number of correctly predicted labels for the optimal feature subsets was 99.35% for RF, 99.02% for KNN, 98.20% for SVM, and 97.61% for LR, whereas the baseline accuracies for the cases with all the features were 95.41%, 94.76%, 92.86%, and 89.55%, respectively. Moreover, the execution times decreased by almost one-third, showing the efficiency of the method. The report indicates validation of the PSO-based fitness function developed for lightweight-accuracy attacks detection in IoMT devices, thus proving to be efficient and cost-conducive, readily deployable in medical organizations as well as smart home environments, thereby safeguarding future-proof healthcare infrastructures against dynamically evolving threats.

As the data volume of astronomical imaging surveys rapidly increases, traditional methods for image anomaly detection, such as visual inspection by human experts, are becoming impractical. We introduce a machine-learning-based approach to detect poor-quality exposures in large imaging surveys, with a focus on the DECam Legacy Survey (DECaLS) in regions of low extinction (i.e., E ( B − V ) < 0.04 ). Our semi-supervised pipeline integrates a vision transformer (ViT), trained via self-supervised learning (SSL), with a k-Nearest Neighbor (kNN) classifier. We train and validate our pipeline using a small set of labeled exposures observed by surveys with the Dark Energy Camera (DECam). A clustering-space analysis of where our pipeline places images labeled in good and bad categories suggests that our approach can efficiently and accurately determine the quality of exposures. Applied to new imaging being reduced for DECaLS Data Release 11, our pipeline identifies 780 problematic exposures, which we subsequently verify through visual inspection. Being highly efficient and adaptable, our method offers a scalable solution for quality control in other large imaging surveys.

Background: Auditory hallucinations (AHs)—perceptions of sound without external stimuli—are common in clinical populations but rarely investigated in healthy individuals. This study aimed to employ Pavlovian conditioning to induce AH-like experiences in healthy subjects and to examine their neural correlations using electroencephalography (EEG). Methods: Seven healthy volunteers were exposed to auditory, non-auditory, and conditioned non-auditory stimuli while recording their EEG with a 32-channel system. Results: When comparing “sound” (auditory) and “conditioned no sound” (conditioned non-auditory) scenarios, the differences in average EEG power were much less pronounced compared to regular sound/no sound scenario. However, significant alterations (p = 0.05) in β and γ rhythms were observed in bilateral temporal regions when comparing the “no sound” and “conditioned no sound” scenario, resembling the spectral patterns observed during real sound perception. These EEG alterations indicate successful induction of hallucination-like auditory experiences through Pavlovian conditioning. A three-class k-nearest neighbor (kNN) classifier detected AH-like events with >80% accuracy in six out of seven participants. Conclusions: Pavlovian conditioning can induce AH-like perceptions in healthy individuals, accompanied by measurable EEG alterations. Therefore, EEG-based methods have the potential for objective detection and assessment of auditory hallucinations and offer a foundation for future research on their neural mechanisms.

Featured data in a dataset can affect the data processing, either for the better or for the worse. In addition, feature data can also affect the time of data processing. Selection of the right feature data may need to be done where the feature data can represent the whole of a dataset. In this study, a search for feature data will be carried out that can result in better data processing. The classification process will be carried out on an Iris dataset with the KNN algorithm. The iris dataset has 4 feature data (Sepal Length, Sepal Width, Petal Length, Petal Width) and the exact feature data variation will be determined in this classification. The dataset will be broken down into 7 variations of data and tested with a comparison of the training data and test data, namely 90:10, 80:20, 70:30, 60:40, 50:50, 40:60, 30:70, 20:80 and 10:90. The KNN algorithm used has parameters with the number of n neighbors 5 and the Minkowski metric. In this study, the highest accuracy value was 96% and the lowest accuracy value was 71%. The highest accuracy value is obtained from the variation of the Petal Length and Petal Width data features while the lowest accuracy value is obtained from the variation of the Sepal Length and Sepal Width data features.

BackgroundThe surgical extraction of impacted mandibular third molars present significant clinical challenges, where accurate preoperative assessment is crucial to mitigate risks such as Inferior Alveolar Nerve injury. Although artificial intelligence shows promise in dental radiology, existing approaches are often limited to binary classification, affected by class imbalance, and lack standardized evaluation protocols, thereby restricting their clinical applicability.MethodsThis study proposes two independent deep learning frameworks for comprehensive analysis of third molar impactions. The first framework is an end-to-end object detection pipeline employing modified YOLOv10 and YOLOv11n architectures enhanced with multihead self-attention. The second framework is a feature-based classification approach, where deep features extracted using ResNet50 and InceptionNetV3 are classified using traditional machine learning algorithms.ResultsValidated on a multinational dataset of 5796 expertly annotated orthopantomograms with high inter-rater agreement (κ = 0.92), the proposed frameworks demonstrated competitive performance. The Fine KNN classifier using ResNet50 features achieved the best classification performance, yielding 97.56% accuracy, 96.07% precision, 96.21% recall, and an F1-score of 96.10%, while InceptionNetV3-based classification achieved 97.33% accuracy with an F1-score of 95.30%. For object detection, YOLOv11n attained a mean average precision of 88.9% (mAP@0.5) and 85.7% (mAP@0.5:0.95), while maintaining substantially lower computational complexity (19.7 vs 28.4 GFLOPs). Ablation experiments confirmed that the integration of multihead self-attention modules and generative adversarial network-based augmentation improved detection performance by 6.4% mean average precision.ConclusionsThe proposed frameworks enable accurate and automated multiclass assessment of third molar impactions, achieving high diagnostic performance while preserving computational efficiency suitable for clinical deployment. This work advances artificial intelligence-assisted surgical planning by providing reliable F1-score-based evaluation, reliable real-time detection, and enhanced preoperative risk stratification in oral and maxillofacial surgery.

Denial-of-service (DoS) attacks pose a major threat to various kinds of computer networks. There are several kinds of networks that are victims of DoS attacks, one of them being the wireless sensor network (WSN). The main objective of this work is to detect such attacks in wireless sensor networks. These networks are susceptible to intrusion attacks because of their fragile defense mechanisms in unattended environments. Thus, a suitable intrusion detection system must be created to optimally detect DoS attacks and prevent them. This work proposes a hybrid technique called Grasshopper Optimization Algorithm-Genetic Algorithm (GOA-GA), which combines the advantages of two metaheuristic algorithms, namely, the Grasshopper Optimization Algorithm and the Genetic Algorithm, to optimize feature selection based on the given WSN dataset. After optimal feature selection and training, the machine learning classification algorithms classify whether the traffic is normal or benign in the form of four types of DoS attacks, namely, Blackhole, Scheduling, Flooding, and Grayhole attacks. The proposed model and algorithms used are further validated and compared based on standard performance metrics. The experiments conducted during the research show that the GOA-GA method, when combined with the KNN classifier, achieves an accuracy of 95.51% and a recall of 95.51%, exhibiting competitive performance relative to recent state-of-the-art approaches while reducing feature dimensionality and computational overhead. These results indicate that the proposed hybrid optimization strategy offers a robust and efficient solution for DoS attack detection in WSNs, contributing to ongoing research in information security.

Proton exchange membrane fuel cells (PEMFCs) are attractive energy sources for clean and efficient power generation; however, their nonlinear characteristics and sensitivity to operating condition variations make maximum power point tracking (MPPT) a challenging control problem. Conventional MPPT techniques often exhibit slow convergence, steady-state oscillations, and degraded performance under dynamic fuel flow variations. This paper proposes a machine learning–driven MPPT control strategy for a PEMFC system integrated with a DC–DC boost converter. The MPPT problem is formulated as a supervised classification task, where machine learning classifiers generate duty-cycle commands to regulate the converter and ensure operation at the maximum power point. A detailed PEMFC–converter model is developed in MATLAB/Simulink-2025b, and a dataset of 3000 labeled samples is generated under varying fuel flow conditions. Several classification algorithms, including decision trees, support vector machines (SVM), k-nearest neighbors (kNN), and ensemble learning methods, are systematically evaluated within an identical simulation framework. Simulation results show that the proposed machine learning-based MPPT controller significantly improves dynamic and steady-state performance. Ensemble Boosted Trees achieve the best overall response with a settling time of approximately 32 ms, peak power overshoot below 4.5%, and steady-state power ripple limited to 1.5%. Quadratic SVM and weighted kNN classifiers also demonstrate stable tracking behavior with power ripple below 2.1%, while overly complex models such as Cubic SVM suffer from large oscillations and reduced accuracy. These results confirm that classification-based machine learning offers an effective, fast, and robust MPPT solution for PEMFC systems under dynamic operating conditions.

The effective care of skin cancer relies on the fine detection of skin lesions. Deep learning techniques are increasingly being used in medical diagnosis, ranging from the classification of skin lesions. Their ability to learn deep discriminative features from dermoscopic images is what makes them popular. In spite of the fact that deep learning approaches learn rich semantically rich information, the approaches currently being taken tend to suffer from poor generalization, high levels of redundancy, and KNN classifiers that assign identical weights to all neighbors. The paper proposes a new approach using machine learning for the classification of skin lesions, entailing deep feature extraction, techniques for dimensionality reduction, and approaches for optimization. Specifically, the ResNet50 architecture using Global Average Pooling for deep feature extraction from dermoscopic images will be employed. The most relevant and non-redundant features are identified through the Minimum Redundancy Maximum Relevance (mRMR) method. mRMR removes irrelevant class information and reduces the feature size considerably. A new approach for the KNN classifier substitutes the fully connected layer of ResNet50. The weights for instance and feature levels are computed with the Genetic Algorithm (GA) and the use of cosine similarity. The proposed approach attains a high accuracy of 90.61% on the classification task for the binary images of skin lesions. The experimental results show that the proposed optimized cosine weighted KNN approach is effective for the diagnosis of skin cancer.

Background/Objectives: Depressive disorder (DD) is a prevalent psychiatric condition often diagnosed through subjective self-reports, which can be time-consuming and lead to inaccurate assessments. To enhance diagnostic precision, integrating Electroencephalography (EEG) with machine learning (ML) has gained attention as an objective approach for DD diagnosis and severity assessment. Methods: We propose an interpretable EEG-based ML framework that integrates optimized functional connectivity features, including Coherence, Phase Lag Index (PLI), and Granger causality, to explore EEG-based functional connectivity patterns in individuals clinically diagnosed with depressive DD and to model symptom severity and cognitive vulnerability. The identified biomarkers provide a promising foundation for developing objective, clinically actionable decision-support tools in psychiatric care. Feature selection was performed using the Neighborhood Component Analysis (NCA) method, and biomarkers were identified through statistical tests. Results: The highest classification performance (97.66% ± 2.05%accuracy, 99.20% ± 1.10% sensitivity, 95.91% ± 4.66% specificity, 98.00% ± 1.02% f1-score, and 0.95 ± 0.48 MCC) was achieved using 21 NCA-selected features with a KNN (K = 9) classifier. The best severity assessment (r2 = 0.89 ± 0.10, MSE = 3.96 ± 17.05) and cognitive impairment prediction (r2 = 0.89 ± 0.06, MSE = 0.23 ± 0.45) were obtained using an ANN regressor with 20 and 17 NCA-selected features, respectively. Conclusions: Our approach outperforms previous EEG-based ML models in DD classification and severity prediction using fewer features. Notably, this is the first study to use EEG connectivity features to predict patients' severity and cognitive impairment in DD. Coherence and PLI values from frontal and temporal pathways across the alpha, beta, and gamma sub-bands may serve as critical biomarkers for DD diagnosis, severity assessment, and prediction of cognitive impairment.