Purpose: The objective is to develop a more effective model that simplifies and accelerates the news classification process using advanced text mining and deep learning (DL) techniques. A distributed framework utilizing Bidirectional Encoder Representations from Transformers (BERT) was developed to classify news headlines. This approach leverages various text mining and DL techniques on a distributed infrastructure, aiming to offer an alternative to traditional news classification methods. Design/methodology/approach: This study focuses on the classification of distinct types of news by analyzing tweets from various news channels. It addresses the limitations of using benchmark datasets for news classification, which often result in models that are impractical for real-world applications. Findings: The framework’s effectiveness was evaluated on a newly proposed dataset and two additional benchmark datasets from the Kaggle repository, assessing the performance of each text mining and classification method across these datasets. The results of this study demonstrate that the proposed strategy significantly outperforms other approaches in terms of accuracy and execution time. This indicates that the distributed framework, coupled with the use of BERT for text analysis, provides a robust solution for analyzing large volumes of data efficiently. The findings also highlight the value of the newly released corpus for further research in news classification and emotion classification, suggesting its potential to facilitate advancements in these areas. Originality/value: This research introduces an innovative distributed framework for news classification that addresses the shortcomings of models trained on benchmark datasets. By utilizing cutting-edge techniques and a novel dataset, the study offers significant improvements in accuracy and processing speed. The release of the corpus represents a valuable contribution to the field, enabling further exploration into news and emotion classification. This work sets a new standard for the analysis of news data, offering practical implications for the development of more effective and efficient news classification systems.

The emergence of diverse content-sharing platforms and social media has rendered the dissemination of fake news and misinformation increasingly widespread. This misinformation can cause extensive confusion and fear throughout the populace. Confronting this dilemma necessitates an effective and accurate approach to identifying misinformation, an intrinsically intricate process. This research introduces an automated and efficient method for detecting false information. We evaluated the efficacy of various machine learning and deep learning models on two separate fake news datasets of differing sizes via holdout cross-validation. Furthermore, we evaluated the efficacy of three distinct word vectorization methods. Additionally, we employed an enhanced weighted voting ensemble model that enhances fake news detection by integrating logistic regression (LR), support vector machine (SVM), gated recurrent unit (GRU), and long short-term memory (LSTM) networks. This method exhibits enhanced performance relative to previous techniques: 98.76% for the PolitiFact dataset and 97.67% for the BuzzFeed dataset. Furthermore, the model outperforms individual components, resulting in superior accuracy, precision, recall, and F1 scores. The enhancements in performance result from the ensemble method’s capacity to use the advantages of each base model, hence providing robust generalization across datasets. Cross-validation was employed to enhance the model’s trustworthiness, validating its capacity to generalize effectively to novel data.

Background/Objectives: Alzheimer’s disease (AD), the most prevalent form of dementia, is characterized by an overall decline in cognitive functioning and represents a major public health crisis. It remains critical to be able to accurately and quickly diagnose patients with AD; however, recent deep learning approaches using MRI data do not provide sample generalization, have high computational requirements, and offer little interpretability. Methods: In this study, we present a new framework called eXplorative ViT-CNN (X-ViTCNN) that combines a customized Vision Transformer model with two previously trained CNNs (DenseNet201 and MobileNetV2). With our proposed preprocessing approach using contrast-enhanced preprocessing to highlight neuroanatomical features as well as Bayesian Optimization to tune hyperparameters, we fuse local structural features originating from the CNNs with global representations from the transformer and feed the final result to fully connected dense layers for multi-stage classification. We also use Grad-CAM visualizations to provide insight into how our model arrived at its classification. Results: Experiments conducted on ADNI and OASIS datasets demonstrate the superiority of X-ViTCNN, achieving accuracies of 97.98% and 94.52%, respectively. The model outperformed individual baselines and other pre-trained architectures, showing balanced sensitivity and specificity across all AD stages. Conclusions: The proposed X-ViTCNN framework is a powerful, interpretable method for predicting the development of multi-stage Alzheimer’s disease using MRI scans. The combination of complementary feature learning, automatic hyperparameter optimization and interpretability through visualization make it an excellent potential tool for clinicians to support their decision making in the early diagnosis and ongoing monitoring of persons with Alzheimer’s disease.

Effective diabetes management requires precise non-invasive monitoring systems that surpass the limitations of traditional glucose testing. Based on photoplethysmography (PPG) signals, this study presents a novel Artificial Intelligence (AI) driven framework for non-invasive blood glucose monitoring (NIBGM) using machine learning, deep learning, and neural network on four diabetic classes (hyperglycaemia, hypoglycaemia, pre-diabetic, and normal) that are further validated by five statistical techniques (10 K-fold, bootstrap P-value, sensitivity vs false negative, brier-score, heatmap). The data is pre-processed and mapped with continuous glucose monitoring (CGM) readings, further balanced by AI-based Synthetic minority oversampling technique (SMOTE) analysis. The 12 features (morphological, statistical, temporal, and spectral) are scrutinised using a set of four statistical methods (L1 regularization, cross validation, hyper parameter tuning, and recursive method). This research also reports a set of latest six classifiers in conjunction, to assess the AI-based accuracy for the first time on diabetic date set. The classifiers include: Ensemble (bilstm + lightgbm), catboost, gradient boosting models, convolution neural networks, random forest. The ensemble and Lightgbm provided highest of 89% accuracy and 0.889 weighted F1 score. The models showed flawless sensitivity for all four classes. The obtained results are validated by a set of five abovementioned statistical techniques concluding an extensive layer of validation, not reported yet. The results conclude the clinical potential of framework to provide accessible, accurate, and thus dependable diabetes monitoring smart healthcare AI technologies.

The early and precise classification of wheat leaf diseases is crucial for safeguarding crop yield and ensuring food security, as timely intervention can prevent significant agricultural losses. This study introduces an advanced deep learning framework that integrates Swin Transformer and VGG-16 to enhance feature extraction and classification performance. The model is evaluated on three benchmark datasets: the Mendeley dataset, the Kaggle wheat disease dataset, demonstrating superior performance across all key metrics. Experimental results show an overall classification accuracy of 97.50% on the proprietary dataset and 92.50% on the Mendeley dataset, with high precision, recall, F1-score, and AUC values across all disease classes. A factual assessment assist approves the strength of the demonstrate, with Cohen's Kappa scores of 0.9674 and 0.8794, affirming solid inter-rater unwavering quality. The proposed approach beats standard models in terms of exactness, strength, and generalization, especially in challenging conditions like changing lighting and natural clamor. Comparative examination against conventional CNN-based models highlights the adequacy of Transformer-based designs in capturing complicated spatial conditions and fine-grained illness designs inside leaf structures. Swin Transformers were chosen over conventional CNNs due to their progressive plan and moved window-based self-attention instrument, which permit for more effective dealing with of long-range conditions and adaptable highlight extraction. The proposed approach not as it were moves forward classification precision but too improves the robotization of illness location, in this manner encouraging effective illness checking in accuracy agribusiness. However, the model's prevalent execution comes with a trade-off in computational fetched, counting higher memory utilization and preparing time. By leveraging self-attention instruments, the show viably centers on disease-specific locales, decreasing misclassification and progressing generally generalization. These discoveries emphasize the potential of progressed profound learning models in changing rural infection determination, empowering adaptable and exact plant infection classification frameworks.

Interpretable detectors built on prosodic and voice-quality measures have been proposed as an alternative to black-box spectral models for audio deepfake detection. We analyze 26 acoustic-prosodic features in a public corpus of 1,866 clips: 933 real clips cut from fourteen YouTube recordings and 933 clips produced by six text-to-speech systems. The corpus contains repeated recordings and exact-duplicate audio, so clips are not independent observations. We therefore recover source identifiers from filenames, merge sources linked by duplicate audio into 93 canonical source groups, and treat the canonical source group, not the clip, as the unit of independence. Source-grouped cross-validation is the primary evaluation; clip-level cross-validation is reported only to measure the optimism that source leakage introduces. All 26 features differed between classes at the clip level, but only ten remained significant after false-discovery-rate correction when each source group contributed a single observation. Ignoring source structure raised the mean per-fold ROC-AUC by 0.033 for logistic regression and 0.044 for random forest. A nested, source-grouped model-selection procedure gave a pooled out-of-fold ROC-AUC of 0.943, an average precision of 0.944, and an equal error rate of 12.2\%. When each real recording was held out in turn, the proportion of its clips classified as real ranged from 0.48 to 1.00, so performance on an unseen real recording depends strongly on the recording. Because all real speech comes from fourteen recordings and recording provenance is confounded with class, these results establish acoustic separability within this corpus only; they do not demonstrate a general ability to distinguish human from synthetic speech.