Search Results (1,183)

This study presents an inappropriate-content classification framework for digital-violence detection in Ecuadorian Spanish, addressing extreme class imbalance and dialectal variation. Data were collected in a post-API setting through a Selenium-based scraping pipeline that reconstructs conversational context using a window of ? = 3 prior interventions. An initial zero-shot labeling attempt with LLMs (Hermes) revealed severe cultural misinterpretation, overestimating the Violence class by 50 times (97.5% false alerts), which motivated full human validation and targeted data engineering. To correct imbalance without contaminating evaluation, the corpus was split before augmentation (70/15/15), and minority classes were selectively leveled via few-shot generation with LLaMA 3.1, followed by strict deduplication and cosine-similarity filtering (? = 0.85) to preserve semantic diversity. Model selection compared BETO and mBERT, with BETO outperforming. Across four training scenarios, naïve oversampling produced artificially inflated metrics indicative of overfitting, whereas the proposed cost-sensitive and regularized configuration (BETO with semantic deduplication and weighted loss) achieved 94.39% accuracy, 0.9429 weighted F1, and 0.9022 macro F1, significantly improving recovery of critical classes. Results highlight that hybrid data are effective only when carefully curated and paired with leakage-free evaluation protocols. This work demonstrates how machine learning innovation and knowledge extraction from heterogeneous data can be combined to build robust models for digital-violence detection in low-resource, culturally specific contexts. Full article

As critical equipment in power systems, the reliable operation of power transformers is directly linked to the overall safety of the power grid. Traditional fault diagnosis methods based on dissolved gas analysis generally rely on a single gas feature, which inevitably causes misjudgment and suffers from inadequate accuracy. This paper proposes a transformer fault diagnosis model that integrates multidimensional features with intelligent algorithms. The model adopts the volume fractions of five key gases from dissolved gas analysis as the fundamental features and further introduces the three-ratio coding features derived from these five gases. The two categories of features are jointly constructed into a multidimensional input vector. In the method design, firstly, the self-adaptive synthetic over-sampling (SASYNO) method is adopted to address the sample imbalance problem; secondly, the least squares support vector machine (LSSVM) is optimized using the experience exchange strategy (EES), and the classification performance of the model is improved. Experimental results demonstrate that the diagnostic model built on multidimensional features significantly outperforms traditional methods in accuracy, thereby providing an effective new approach for transformer fault diagnosis. Full article

Background: Dengue fever remains a major global health problem, and early diagnosis is challenging where confirmatory testing is limited. Machine-learning studies using routine hematological data have focused mainly on discrimination, whereas calibration, decision-analytic performance, interpretability, and robust validation have received less attention. This study aimed to develop and compare tree-based machine-learning models for dengue classification, benchmark them against L2-penalized logistic regression (LR), and evaluate discrimination, calibration, potential decision-analytic benefit, and interpretability. Methods: This retrospective secondary analysis used an open-access dataset from Bangladesh comprising 1523 patients, 18 demographic and hematological predictors, and a binary dengue test outcome. Data were divided into stratified training (80%) and test (20%) sets. The Synthetic Minority Over-sampling Technique was applied only within the training workflow. Random Forest (RF), XGBoost, and LightGBM were optimized using Optuna with stratified five-fold cross-validation. L2-penalized LR was evaluated using the same predictors and training–test partition. Held-out test-set performance was assessed using AUROC, AUPRC, accuracy, sensitivity, specificity, predictive values, F1-score, and Brier score. Calibration, decision curve analysis, SHAP values, and permutation importance were also examined. Results: LightGBM, RF, and XGBoost yielded AUROCs of 0.709, 0.704, and 0.702, respectively, indicating closely similar discrimination. The primary SMOTE-trained LR yielded a numerically lower AUROC of 0.608 (95% CI: 0.536–0.677) and a higher Brier score of 0.310 than the tree-based models (0.174–0.176); however, in sensitivity analysis without SMOTE, the LR AUROC increased numerically to 0.655 and the Brier score decreased to 0.194. At the training-derived threshold of 0.558, LightGBM achieved a sensitivity of 0.914 and a specificity of 0.458, reflecting a high-sensitivity, low-specificity profile. The LightGBM calibration curve suggested closer agreement in the low-to-moderate predicted-probability range, with greater deviation at higher probabilities. Decision curve analysis suggested potential net benefit across a range of threshold probabilities but did not establish clinical utility. Platelet count, monocyte percentage, and neutrophil percentage were consistently among the leading predictors across the tree-based models. Conclusions: Tree-based models showed moderate discrimination, with high sensitivity but limited specificity, and yielded numerically higher AUROCs and lower Brier scores than the primary SMOTE-trained LR within this internal-validation framework. They should not replace etiological testing or be used as standalone diagnostic tools; their observed operating characteristics are more compatible with a potential adjunctive screening or triage-support role. External and prospective validation across independent populations and settings is required before clinical use or superiority over simpler statistical models can be established. Full article

Early stroke risk prediction offers an opportunity for timely interventions and may help reduce the clinical burden associated with stroke. Artificial intelligence (AI) provides medical practitioners with tools to analyse clinical biomarkers and predict a patient’s stroke risk. However, existing models lack interpretability and explainable decision support, limiting their adoption in clinical settings. This paper proposes a hybrid Gated Recurrent Unit (GRU)-TabTransformer architecture using cross-attention for stroke-status prediction. The proposed architecture comprises two stages. In the first stage (Model A), ordered feature-sequence representations from a GRU encoder are combined with concatenated categorical and numerical tabular features from a TabTransformer encoder. The model passes these distinct learned representations through cross-attention and linear projection layers before the final prediction. In the second stage (Model B), we augment our model with Large Language Models (LLMs) and Local Interpretable Model-agnostic Explanations (LIME) to provide per-sample, post hoc, human-interpretable explanations based on predicted probabilities. Experimental results show that both models achieve competitive sensitivity and accuracy values. In particular, the Synthetic Minority Over-sampling Technique (SMOTE) yielded a more balanced sensitivity–specificity trade-off, with a sensitivity of 74.00% and a specificity of 75.10%. Moreover, the model achieved an accuracy of 75.05% with SMOTE. An ablation study further shows that Cross-Attention offers better sensitivity and Receiver Operating Characteristic-Area Under the Curve (ROC-AUC), while Gated Fusion performs better on several other metrics. Additionally, age and average glucose level were the most influential stroke risk indicators, while Body Mass Index (BMI) and ever-married status were secondary model-attributed features. A two-factor repeated-measures Analysis of Variance (ANOVA) confirmed an interaction between model choice and the class-balancing technique used in stroke risk prediction systems. The Mistral + Hybrid GRU-TabTransformer architecture also recorded a mean inference time of 57.59s using few-shot prompting. Overall, the results provide a proof of concept for integrating hybrid GRU-TabTransformer with cross-attention and LLM-based explainability to support interpretable stroke risk prediction systems, pending robust external validation before deployment. Full article

While debate persists over the accuracy of retrospective self-reports of childhood family environments, empirical understanding of the stability of these assessments among young adults during stressful events remains limited. In response, this study examined whether college students’ perceptions of problematic childhood family environments changed across the COVID-19 onset. We leveraged a longitudinal dataset of 355 college students (237 female, Mage = 19.4 years, SD = 0.70; 118 male, Mage = 19.6 years, SD = 0.70), oversampled for recent substance misuse. Participants completed assessments over five to seven timepoints; the pandemic onset (13 March 2020) occurred near the midpoint of data collection. We employed piecewise growth-curve modeling (PGCM) to evaluate pandemic-onset-related changes in participants’ perceptions of problematic childhood family environments and the moderating effect of family co-residence during the acute pandemic onset. Results indicated that college students’ perceptions of problematic childhood family environments significantly but modestly increased from pre- to post-pandemic onset. Living with family during the acute disruption period did not moderate this change. In conclusion, stressful events like the COVID-19 outbreak could serve as a pivotal event associated with increased problematic perceptions of childhood family environments among young adults. We reinforce the necessity of adopting rigorous longitudinal, repeated-measures designs for retrospective assessments in family sciences. Additional conceptual and theoretical, methodological, and practical implications were discussed. Full article

This paper presents a machine learning-based decision support system for multiclass defect detection, utilizing exclusively heterogeneous legacy process data to minimize manual inspection time in foundries. Validated on 51,377 products across 193 defect categories, the methodology resolves structural data inconsistencies through k-nearest neighbor (kNN) imputation and piecewise winsorization. A multi-stage feature selection cascade, incorporating variance thresholding, correlation filtering, and Random Forest Feature Importance (RFFI), reduces the feature space from 139 to 78 process-critical variables. Following Synthetic Minority Over-sampling Technique (SMOTE)-based class balancing, five classifiers were benchmarked via 10-fold cross-validation and optimized using the macro-averaged F₃-score to mathematically penalize undetected defects. Light Gradient Boosting Machine (LightGBM) and Random Forest (RF) provided superior predictive baselines. To enforce strict zero-defect constraints, an asymmetric risk function shifted decision boundaries, enabling the risk-calibrated LightGBM model to reduce manual inspection volume by 9.72% with zero defect escapes. For resolving conflicting predictions, multi-algorithm decision fusion was implemented. By statistically evaluating the joint probabilities of the base models’ post-calibration outputs, a Naive Bayes Stacking meta-classifier effectively neutralizes single-algorithm inductive biases. Ultimately, synthesizing these F₃-optimized, risk-calibrated base models via meta-learning successfully isolated true defect-free components, maximizing the final inspection time reduction to 14.82% while strictly maintaining zero defect escapes. Full article

Background: Class imbalance remains a major limitation in artificial intelligence based oral cancer diagnosis, particularly in small clinical image datasets where malignant lesions are underrepresented. Methods: Using a dataset of 3000 mobile-acquired oral cavity images categorized into four diagnostic classes (healthy, benign, oral potentially malignant disorders, and oral cancer), we systematically evaluated the impact of imbalance mitigation strategies on diagnostic performance. Three backbone architectures, EfficientNet, Vision Transformer, and Swin Transformer, were trained under four conditions: raw imbalanced training, random under-sampling, random over-sampling, and medical-safe data augmentation. Performance was evaluated using stratified five-fold cross-validation, and all reported metrics correspond to the mean performance across the validation folds. Results: In multiclass evaluation, augmentation yielded the strongest overall performance, with EfficientNet achieving the highest macro-${\mathrm{F}}_1$ score (0.669 ± 0.024) and area under curve (0.881 ± 0.015). Under clinically oriented binary malignant-risk evaluation (high-risk vs. low-risk lesions), sensitivity reached 0.817 ± 0.030 with EfficientNet, while data augmentation maintained competitive malignant-risk detection across the evaluated architectures. In contrast, random undersampling consistently yielded lower descriptive performance across architectures. Conclusions: These findings indicate that augmentation-based imbalance mitigation is associated with improved multiclass and oral cancer specific performance while maintaining competitive overall discrimination. The results provide practical guidance for developing clinically robust artificial intelligence systems for early oral cancer detection in imbalanced real-world datasets. Full article

Background: Early and accurate detection of depression from multimodal data is a critical yet challenging task. Many existing models either rely on a single modality or exhibit limited robustness across datasets, while publicly available datasets are often small and highly imbalanced. Methods: In this study, we propose a robust multimodal fusion framework that leverages bidirectional cross-modal attention to effectively integrate audio and text features, allowing the model to learn complementary information from both modalities. To address class imbalance and limited data, we employ SMOTE-based oversampling. Our model further incorporates validation-based adaptive thresholding and probability ensembling to enhance decision robustness and generalization. Experiments are conducted on two benchmark datasets, EATD-Corpus (Chinese; 162 participants) and DAIC-WOZ (English; 189 usable interviews), and ablation experiments compare cross-modal attention settings. Results: On EATD-Corpus, our fusion model achieves strong performance (F1: 0.820, Recall: 0.934, Precision: 0.762), achieving higher mean performance than the text-only and audio-only baselines evaluated under the same participant-level five-fold protocol. On DAIC-WOZ, the framework remains competitive (F1: 0.836, Recall: 0.925, Precision: 0.775). The ablation results confirm that bidirectional cross-modal interaction yields the most reliable trade-off between Precision and Recall. Conclusions: This paper proposes a deep learning framework for depression screening that jointly leverages audio and text modalities, demonstrating robust performance on small, class-imbalanced datasets and architectural generalizability across two corpora with different languages and interview settings. Full article

Tractography from diffusion MRI enables the mapping of white matter pathways. A key challenge is the automatic segmentation of streamlines into anatomically meaningful bundles, which is strongly affected by class imbalance: common bundles dominate the training data, whereas bundles such as the fornix are severely underrepresented (<0.2% of streamlines). This work proposes a tractography segmentation framework that combines a generative oversampling strategy based on Variational Autoencoders (VAEs) with a streamline-wise convolutional classifier operating on FiberMap representations. The generative model is applied in a targeted manner only to underrepresented bundles, and its effect is compared with two alternative oversampling schemes: simple sample duplication and the Synthetic Minority Oversampling Technique (SMOTE). Experiments on the Tractoinferno dataset, comprising 32 white matter bundles, show that all configurations yield similar global performance, with macro F1-scores around 90–91%. In contrast, performance on underrepresented bundles changes markedly, with VAE-based oversampling improving left/right fornix F1-scores from 77.2%/68.7% to 90.4%/92.5%, yielding an average gain of 18.5 percentage points across the two fornix bundles. Sample duplication increases recall but substantially reduces precision, while SMOTE achieves the highest recall at the cost of more false positives. A complementary three-dimensional geometric analysis revealed a fidelity–coverage trade-off, with SMOTE more closely reproducing real length and curvature distributions while VAE samples provided greater coverage of real trajectories. Overall, the VAE achieved the most favorable precision–recall balance for the targeted fornix bundles. These results suggest that targeted VAE-based oversampling is a practical strategy for mitigating extreme imbalance in tractography segmentation, improving the recognition of underrepresented bundles without materially affecting performance on dominant bundles. Full article

In current state-of-the-art intrusion detection systems (IDSs), models are trained to detect the repetitive behavior of data traffic. This leads to the problem of ignoring rare yet important attacks. To solve the problem stated above, a risk-gated security attention (RGSA) architecture is proposed. The initialization of the risk estimator based on the CVE/CVSS severity score enables the attention network to detect important yet rare patterns. Furthermore, a tier-aware focal loss is proposed to mitigate security threats in the real world without any data augmentation process. Evaluations were conducted on held-out 80/20 test splits of the CIC-IDS2017, CIC-IDS2018, and CIC-IoT2023 datasets. The calibrated binary detection process reached false negative rates (FNRs) of 0.50% and 0.40% on the CIC-IDS2017 and CIC-IDS2018 datasets, respectively, remaining below the 1.0% FNR operational target adopted in this study. For micro-support critical threats, precise discrete analysis had to be applied to maintain statistical validity. Also, weighted F1-scores of at least 99.65% were consistently achieved for all testbeds analyzed. This network contains 284,317 trainable parameters and strictly avoids the creation of any synthetic samples through complete rejection of synthetic oversampling. The findings indicate that adding priors for security tiers to the attention process is favorable for moving towards an impact-based threat management approach. Full article

Class imbalance is prevalent in real-world datasets. Minority samples are far fewer than majority samples. Traditional classifier design typically assumes balanced data, which causes classifiers to favor the majority class when faced with imbalanced datasets. Thus, there are high misclassification costs for minority classes in critical fields like healthcare and finance. Most existing oversampling methods for handling imbalance problems, such as SMOTE (the synthetic minority oversampling technique), suffer from limitations like noise sensitivity, failure to consider minority-class sub-cluster structures, and poor adaptability to the heterogeneity of sample distributions. This article addresses these issues by proposing a novel oversampling algorithm: the adaptive weighting–synthetic minority oversampling technique (AW-SMOTE). It consists of three progressive stages. In the first stage, DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering identifies the distribution structure of the minority samples. It identifies potential sub-clusters while removing noise interference. A clear data foundation for subsequent sampling is established. In the second stage, the most representative boundary sample in each cluster is used to evaluate the weight of each cluster. The total number of synthetic samples is allocated to different clusters according to their weights. This provides global sample enhancement support. In the third stage, adaptive sample generation is performed within each cluster. It combines the two perspectives of boundary tightness and local density. The sigmoid function is used to dynamically adjust the weight ratio. Finally, new samples are synthesized in key regions to both preserve distribution characteristics and enhance discriminability in classification. Through experiments on standard datasets from the KEEL repository, the feasibility and effectiveness of this algorithm are demonstrated. Full article

Timely and accurate classification of lung diseases from chest X-ray images remains an important healthcare challenge. Machine-learning and deep-learning methods can support automated classification, but their evaluation may be affected by class imbalance, feature redundancy, dataset leakage, and computational cost. This paper evaluates an artificial immune system (AIS)-evolved decision-tree ensemble using fold-specific ResNet18 features. All within-dataset experiments use duplicate-family-aware five-fold splits. Within each fold, standardization and adaptive principal component analysis (PCA) are fitted to the training features, and the Synthetic Minority Over-sampling Technique (SMOTE) is applied only to the reduced training data. Each candidate tree is assigned an affinity based on out-of-bag macro-F1. In the primary run, mean within-dataset macro-F1 was 98.07%, 99.48%, and 98.26% for Datasets 1–3, respectively, and 96.66% for the exploratory Dataset 4. Because of extensive cross-dataset image reuse and conflicting labels, Dataset 4 does not provide independent evidence of clinical lung-cancer detection. A five-seed repeated-initialization analysis repeated the complete fold-specific feature and classification pipeline while preserving the same folds. Mean macro-F1 differences between AIS and the prespecified static comparator for each dataset, calculated as AIS minus the comparator, were $-0.18$, 0.00, $-0.35$, and $-0.13$ percentage points for Datasets 1–4, respectively. Using the same sign convention, mean differences between AIS and the fixed random tree ensemble ranged from $-0.05$ to $+0.05$ percentage points. Population diagnostics showed that evolution improved individual-tree macro-F1 but reduced pairwise disagreement, without a consistent majority-vote gain. Median latency from an already decoded image to prediction ranged from 38.86 to 61.20 ms on one CPU thread and from 3.43 to 6.36 ms on an RTX 4090. The results do not establish a practically important or consistent predictive advantage from AIS evolution. The study provides a reproducible and duplicate-controlled framework for evaluating AIS-based tree ensembles. Full article

Identifying the small group of customers who hold a given policy in severely imbalanced tabular data is a recurring screening problem in insurance analytics. This study considers binary caravan-insurance policy-ownership classification on the COIL 2000 benchmark, where the positive-class prevalence is below 6%. The benchmark is a single cross-section, so the label describes current ownership rather than a future purchase event. We propose an Attention-based Symmetric AutoEncoder (ASAE) that combines an auxiliary symmetric reconstruction branch, a channel attention gate, and a marginal log-variance regularizer on a 32-dimensional latent representation. The regularizer operates on individual latent variances and is treated as a heuristic rather than as an estimator of joint differential entropy. Under a common tuning and evaluation protocol on a stratified partition, the ASAE is compared with five conventional machine learning methods and seven neural models. Across five paired runs, it achieves an F1-score of 0.6008 ± 0.0115 and an area under the receiver operating characteristic curve (AUC) of 0.8584 ± 0.0034. Relative to TabNet, the strongest baseline considered, the mean differences are 0.064 in F1-score (95% confidence interval 0.043–0.085) and 0.032 in AUC (95% confidence interval 0.017–0.048). The ordering is preserved across five stratified re-splits, four imbalance-handling configurations, and a complete type-consistent preprocessing rerun in which nominal attributes are one-hot encoded, oversampled with SMOTENC, and reconstructed with categorical cross-entropy losses (F1-score 0.6241 ± 0.0074, AUC 0.8702 ± 0.0050). All reported results use stratified random partitions of COIL 2000. Because 27% of the records share an identical predictor vector with another record, the official challenge separation and two grouped partitions are also defined, so that exact-duplicate and sociodemographic overlap can be isolated from the primary split. The training code, split indices, and per-run predictions used for the reported tables are publicly available. Full article

Forest fires remain a major threat to biodiversity, human settlements, and the climate. This study presents Nuthur, an intelligent wildfire prediction and early-warning system for Al-Soudah, Saudi Arabia, integrating near-real-time environmental data with artificial intelligence models. The system used the Algerian Forest Fire dataset and a newly created local Saudi Arabian dataset. L1 regularization and Recursive Feature Elimination with Cross-Validation (RFECV) were used to examine relevant environmental variables, while oversampling, undersampling, and (Conditional Tabular Generative Adversarial Network) CTGAN-based synthetic augmentation were evaluated to address class imbalance. Multiple ML and DL models were evaluated, including Random Forest (RF), SVM, XGBoost, CatBoost, ensemble models, MLP, TabNet, and exploratory LSTM and CNN models, which were not interpreted as temporal or spatial models. Under random five-fold cross-validation, ML models achieved accuracy values from 0.89 to 0.97, with XGBoost with oversampling achieving the highest accuracy of 0.97. Deep-learning models achieved accuracy values from 0.75 to 0.94, with TabNet using RFECV achieving the best deep-learning result. A separate spatial cross-validation analysis of the Saudi dataset showed lower geographic generalization performance. CatBoost without oversampling achieved the highest mean spatial accuracy (0.8317) and weighted F1-score (0.7782), while logistic regression with oversampling achieved the highest fire-class recall (0.5616). In contrast, XGBoost with oversampling had a fire-class recall of 0.1096. These results highlight the need for further geographically and temporally diverse Saudi data before operational deployment. Full article

Digital fluxgate magnetometers have been widely used in deep space exploration due to their low noise, high sensitivity, and high reliability. This paper presents a digital fluxgate magnetometer using a fifth-order single-loop 1-bit Sigma–Delta modulator. With a consistent system structure, measurement range, test setup, and calculation method, the characteristics of magnetic field measurement noise and non-linear error are obtained under four OSR configurations through simulation analysis and experimental testing. The test results show that within the range of ±65,000 nT, the system achieves its optimal performance with a non-linearity of 0.024%, an RMS noise of 0.106 nT, and a noise power spectral density of 5.7 pT·Hz^−1/2 at 1 Hz. These results indicate that increasing the OSR can effectively improve the performance of this digital fluxgate magnetometer, enabling high linearity and low noise measurement in Earth’s magnetic field. Full article