Electrocardiogram (ECG) signal classification is essential for identifying cardiovascular disorders, and the utilization of deep learning methodologies has demonstrated potential in improving classification precision. In this paper, we evaluate the efficacy of deep learning architectures in categorizing 12-lead ECG signals into distinct cardiac states, utilizing data derived from the CPSP 2018. We pre-processed the ECG signals to normalize the data and employed multiple deep learning architectures, including ResNet 34, to categorize the ECG signals into predetermined categories. The efficacy of these models was assessed using criteria including accuracy, sensitivity, and specificity. The model achieved an overall average F1 score of 80

Smart contracts, as fundamental components of transactions in blockchain and decentralized systems, have inherent risks due to their immutable code and security vulnerabilities that are often challenging to detect. Traditional static analysis tools may overlook certain vulnerabilities, prompting the need for enhanced detection methods. To address this limitation, we propose a hybrid approach that combines static analysis with machine learning techniques. By leveraging static tools, we extract graph-based features from smart contracts such as the number of edges, functions, variables and cycle and we use them to train machine learning models like XGBoost and Random Forest. Our experimental results show that Random Forest demonstrated particularly strong accuracy, showing significant improvements over traditional vulnerability detection methods.

Breast cancer remains one of the leading causes of death among women worldwide, making early and reliable screening a clinical priority. In this work, we propose a lightweight computer-aided diagnosis framework for classifying infrared thermograms based on MobileViT and transfer learning. The pipeline begins with the preprocessing of the thermal images, including normalization, resizing, and edge detection using the Canny and Sobel operators, to construct a three-channel discriminative representation. These data are then used to train a model tailored for a binary classification task (healthy/sick). The experiments were conducted on the DMR-IR dataset, which contains thermal images of both normal and pathological cases. The training process involves fine-tuning a pre-trained model, enabling improved feature extraction in data-scarce settings while maintaining good computational efficiency. The results obtained highlight the effectiveness of this approach and confirm the value of combining lightweight architectures, transfer learning, and contour-based preprocessing for thermographic analysis. This framework thus emerges as a promising solution for screening in resource-constrained environments.

This work presents an innovative methodology for identifying instances of honey adulteration by utilizing the Vision Transformer (ViT) model and thermal imaging techniques to assess and classify honey samples. Conventional techniques employed for the identification of honey adulteration are characterized by protracted processing durations and frequently exhibit limited sensitivity. Thermal imaging is a distinctive benefit as it enables the identification of temperature fluctuations within honey samples, hence facilitating the assessment of disparities in sugar composition, moisture levels, and the existence of adulterants. Thermal imaging technique offers a notable advantage in the detection of adulterants, as it may reveal temperature variations within honey samples caused by differences in sugar composition, moisture levels, and other adulterating substances. To establish a dependable method for classifying honey, we gathered an extensive dataset comprising thermal pictures of 9 unadulterated honey samples, as well as 84 honey samples that were contaminated at varying levels ranging from 1

Diabetic retinopathy (DR), a progressive microvascular complication of diabetes, remains a leading cause of preventable blindness worldwide. Its clinical progression—from mild non-proliferative to severe proliferative stages—demands early and accurate detection to enable timely therapeutic intervention. Traditional diagnostic practices rely on the manual analysis of retinal fundus images, yet these approaches face significant limitations due to the scarcity of ophthalmic specialists and restricted access to screening, particularly in low-resource settings. This underscores the critical need for automated, scalable, and high-performing diagnostic solutions. In response, this study introduces SCGNet, an advanced deep learning framework for automated DR classification that integrates graph-based relational reasoning, residual convolutional architectures, and a meta-learning ensemble strategy. The system is supported by a comprehensive preprocessing pipeline featuring adaptive Master-Slave artifact filtering and progressive image enhancement to optimize visual feature quality. SCGNet was trained exclusively on the APTOS 2019 dataset and rigorously evaluated on two external datasets—IDRiD and a clinically curated private dataset—to assess its generalizability across diverse populations and imaging conditions. On APTOS, the model achieved an accuracy of 99.45

The analysis of data and information is critical, and for stock market traders or investors this can be very crucial. Especially with the progress the world has seen with technology in recent years, as these improvement and changes may come to affect how people perceive things and how they take decision. In this paper we are going to use two methods to classify tweets of Tesla stock that ranges from 30-09-2021 to 30-09-2022, preprocess the features along with financial data from Yahoo Finance to predict close price of the following day using LSTM. Therefore comparing the price prediction models resulted from each method of sentiment analysis, and observing the role of sentiment extraction models in determining the accuracy of price prediction models.

Diabetic Retinopathy (DR) is a vision-threatening complication of diabetes, requiring early detection to prevent irreversible damage. This paper presents an automated DR classification system leveraging advanced preprocessing and deep learning to address key challenges in fundus image analysis. Our approach combines Contrast- Limited Adaptive Histogram Equalization (CLAHE) for contrast enhancement, marker-controlled watershed transformation for optic disc removal, and gradient-based filtering for blood vessel suppression. A federated fuzzy K- means algorithm segments retinal regions, improving feature extraction. The preprocessed images are classified using a fine-tuned InceptionV3 model optimized with mixed-precision training and Nadam, while class-weighted loss mitigates data imbalance. Evaluated on the APTOS 2019 dataset, our model achieves a weighted precision of 93.8

This paper investigates the use of Vision Transformer models for the diagnosis of cancer-related skin lesions. The study focuses on six types of lesions: actinic keratosis (ACK), basal cell carcinoma (BCC), melanoma (MEL), nevus (NEV), squamous cell carcinoma (SCC), and seborrheic keratosis (SEK). The dataset employed included 2,298 smartphone photos of various resolutions, which provided a robust foundation for training and validation the model. To prepare the data, pre-processing techniques including rotation, zooming, and flipping were applied, increasing the minority classes and improving the diversity and robustness of the training set. The model was evaluated in terms of performance metrics, including an accuracy of 76

Breast cancer is a worldwide health crisis that affects a large number of women. Early detection of this disease is critical for determining effective treatment options and improving the chances of positive patient outcomes. In our study, we introduce a new method for detecting breast cancer in its earliest stages using thermographic images and a compact model that can be easily implemented on smartphones. This method is especially useful in areas where medical resources are scarce. We used multiple edge detection techniques like Canny, Roberts and Sobel, and evaluated their effectiveness to improve the accuracy of our model. Our model, which combines MobileNet V2 with a spatial attention mechanism, outperformed other deep learning networks like Inception ResNet and DenseNet121. Furthermore, with an accuracy rate of 98.88%, our proposed model outperformed current state-of-the-art algorithms. These findings point to the potential of our approach for early breast cancer detection and its practical application in resource-limited settings.

Cardiovascular disease (CVD) continues to be a global public health challenge, making scalable and reliable diagnostics necessary. Electrocardiogram (ECG) is a cornerstone in the assessment of the heart, but its conventional 12-lead configuration has logistical and economic hurdles to initiate in low-resource settings. We present, to oppose this, a deep learning model for automated ECG classification using a sparse three-lead configuration (Leads I, II, and V2), selected because of their outstanding diagnostic yield. With the CPSP 2018 dataset, we employed a clean preprocessing pipeline to handle noise and baseline drift and thereafter trained on a ResNet-34 convolutional neural network to predict nine distinct cardiac conditions. We achieved excellent class-specific F1 scores of 0.90 for atrial fibrillation (AF), incomplete atrioventricular block (IAVB), left bundle branch block (LBBB), and right bundle branch block (RBBB), which indicate the strength of the model in detection of life-threatening arrhythmias and conduction defects. Our experiments demonstrate that a low-lead setup, combined with a robust deep learning model, can offer robust diagnostic accuracy. With better accessibility without compromising accuracy, our approach presents an appealing solution to enable scalable ECG screening at scale in resource-limited healthcare environments.

Diabetic retinopathy (DR) remains a leading cause of preventable blindness worldwide, creating an urgent need for reliable automated screening tools. Diabetic retinopathy (DR) grading of fundus images remains challenging due to fine interclass variations and unbalanced datasets. In this work, we propose Contour-Attentive DenseNet, a new deep learning model, which integrates contour processing attention into DenseNet-169 for improved DR classification. Our model introduces three key innovations: (1) A dual-attention contour processing block, which combines channel and spatial attention to enhance microaneurysm and hemorrhage visibility, (2) Inverse frequency weighted adaptive feature fusion between attention-enhanced low-level features and DenseNet's hierarchical representations, and (3) Class-balanced inverse frequency weighting optimization. On APTOS 2019 dataset, Contour-Attentive DenseNet is shown to provide state-of-the-art performance with quadratic weighted kappa score of 0.829 and F1-measure of 0.836, outperforming baseline DenseNet by 6.2% in kappa without compromising the computational efficiency. Ablation experiments validate the role of the attention module, improving sensitivity for Class 1-2 early-stage DR by 11.3%. Clinical interpretability of the model is improved with attention visualization that reveals it's focused on medically relevant retinal regions. Such advances have significant potential for DR screening automation in resource-limited settings. This work provides both a performance benchmark and a framework for developing clinically reliable DR screening tools that balance efficiency with patient safety.

Honey adulteration poses a huge challenge with considerable health and economic consequences, underscoring the necessity for effective and precise quality evaluation techniques. This research introduces a novel approach for classifying levels of honey adulteration through thermal imaging and Artificial Intelligence (AI). Traditional detection methods are frequently marked by protracted processing durations, elevated expenses, and restricted sensitivity. To mitigate these constraints, a dataset of thermal images was compiled from 15 pure honey samples and 69 adulterated samples including glucose syrup at amounts between 1% and 30%. An adaptable AI model was created to categorize various honey types, attaining elevated accuracy, sensitivity, and specificity across different levels of adulteration. The model achieved a precision and specificity of 100% for pure honey and 1% adulteration, demonstrating strong performance at higher adulteration levels (0.98 and 0.97 for 3% and 5% adulteration, respectively). This methodology offers significant benefits, such as swift identification and versatility across various honey varieties. The results indicate that the integration of thermal imaging and AI can improve quality control in the honey sector, providing a dependable method for verifying the authenticity and safety of natural bee products. This approach facilitates enhanced quality assurance methods and bolsters consumer confidence in honey products.

The research suggests a novel approach for the determination of honey adulteration through the Vision Transformer (ViT) model and thermography techniques in honey sample analysis and grading. Honey adulteration, being a natural and highly prized dietetic food item, is a major economic and health hazard. Conventional procedures for the determination of honey adulteration require long processing time and are not as sensitive. Thermal imaging constitutes a distinctive benefit in that it permits temperature difference determination among honey samples and hence facilitates the determination of differences in sugar, moisture, and adulterants. Thermal imaging technique offers a great advantage in adulterant detection since it has the potential to detect temperature variations in honey samples as a result of differences in sugar content, moisture levels, and other adulterants. To find a trustworthy method for honey classification, we gathered a large dataset of thermal images of 9 pure honey samples, and 45 honey samples adulterated at various levels ranging from 1% to 20% during their cooling processes. The dataset was employed to train and fine-tune the model in this work. The findings indicated that the model achieved a level of accuracy at 99.9% with sensitivity of 99.5% and specificity of 100%. The finding of the current study provides the proof to establish the effectiveness of thermal image analysis using Transformers as a capable instrument for the prompt and accurate detection of instances of honey adulteration. The above-mentioned approach presents a likely useful method of implementing quality control policies in the honey industry such that authenticity and safety of this valuable organic resource is ensured.

Skin cancer is a significant global public health issue, with millions of new cases identified each year. Recent breakthroughs in artificial intelligence, especially deep learning, possess considerable potential to enhance the accuracy and efficiency of screening. This study proposes an approach that employs smartphone images, which are preprocessed using adaptive learning and Black-Hat transformation. ViT is utilized for feature extraction, and a stacking model is constructed employing these features in conjunction with image-related variables, like patient age and sex, for final classification. The model's efficacy in identifying cancer-associated skin diseases was evaluated across six categories of skin lesions: actinic keratosis, basal cell carcinoma, melanoma, nevus, squamous cell carcinoma, and seborrheic keratosis. The suggested model attained an overall accuracy of 97.61%, with a PVV of 96.88%, a recall of 97.63%, and an F1 score of 97.19%, so illustrating its efficacy in detecting malignant skin lesions. This method could greatly aid dermatologists by enhancing diagnostic sensitivity and specificity, reducing delays in identifying the most suspicious lesions, and ultimately reaching more patients in need of timely screenings and patient care, thus saving lives.

The swift proliferation and extensive incorporation of the Internet into worldwide networks have rendered the utilization of Intrusion Detection Systems (IDS) essential for preserving network security. Nonetheless, Intrusion Detection Systems have considerable difficulties, especially in precisely identifying attacks from minority classes. Current methodologies in the literature predominantly adhere to one of two strategies: either disregarding minority classes or use resampling techniques to equilibrate class distributions. Nonetheless, these methods may constrain overall system efficacy. This research utilizes Shapley Additive Explanations (SHAP) for feature selection with Recursive Feature Elimination with Cross-Validation (RFECV), employing XGBoost as the classifier. The model attained precision, recall, and F1-scores of 0.8095, 0.8293, and 0.8193, respectively, signifying improved identification of minority class attacks, namely "worms," within the UNSW NB15 dataset. To enhance the validation of the proposed approach, we utilized the CICIDS2019 and CICIoT2023 datasets, with findings affirming its efficacy in detecting and classifying minority class attacks.

During trading sessions, the stock market experiences a multitude of fluctuations that are influenced by an extensive array of factors. This is especially pertinent to businesses that attract substantial international interest, such as streaming platforms that have shown a noteworthy increase in popularity in recent times. The objective of this research is to examine the fluctuations in the value of Netflix shares during the time period that correlates with the release of particular films that elicited significant reactions. The research utilizes natural language processing (NLP) and sentiment analysis approaches to preprocess the movie evaluations and calculate the sentiment score. The study yields thought-provoking findings concerning the correlation between movie release dates and the closing price.

The frequency of cyberattacks increases as network technology evolves and Internet services become more widely utilized. As more individuals and devices connect to the Internet, a substantial amount of traffic data is generated, which facilitates fraudulent activities on these networks. It is essential to have a system that can analyze traffic to prevent criminal activity. Introducing multifaceted IDS approaches to address privacy concerns and security threats using in-depth learning techniques. The performance of deep learning algorithms is highly dependent on the size of the data set and the type of information it contains. Using an unbalanced NSL-KDD data set and model construction steps, we have studied in this article various reduction features and deep learning techniques. LSTM, BiLSTM, and Stacked LSTM are applied to three characteristic reduction methods: Shap values, Boruta, and Anova F-test, in order to compare their results and determine the optimal pair of features to use for the intrusion detection system.

Coronary artery disease is a nuanced cardiovascular condition that is intricate and multifactorial, denoted by the constriction of coronary arteries resulting in compromised blood supply to the cardiac muscle. It necessitates a thorough strategy for diagnosis, management, and prevention, emphasizing lifestyle adjustments and medicinal measures to alleviate the impact of the condition and enhance the well-being of patients. As a result, there is an urgent imperative for the development of cost-effective, automated diagnostic technologies geared towards early CAD detection. Such advancements are crucial for enabling proactive management of chronic cardiovascular conditions within healthcare systems. The rise of machine learning (ML) applications in the medical domain holds immense promise, offering the capability to unravel complex patterns within extensive and diverse medical datasets. By integrating ML methodologies for CAD classification, there lies the potential to alleviate diagnostic uncertainties and improve clinical decision-making processes. This research endeavors to construct a machine learning-driven system tailored specifically for CAD detection. Through meticulous analysis of patient medical records, ML algorithms aim to forecast the likelihood of CAD development in individuals, thereby facilitating timely interventions and risk mitigation strategies. Furthermore, this study seeks to elucidate the pivotal risk factors underlying CAD manifestation, thereby enhancing our comprehension of disease etiology and informing targeted preventive measures. To achieve these objectives, a prominent DL classifier, namely CNN, is deployed to discern patterns within the Z-Alizadeh Sani dataset—a representative repository of real-world patient health records. To address inherent data imbalances, the VAE algorithm is employed to bolster minority class instances. Additionally, model hyperparameters are optimized using the grid search hyperparameter tuning. The resulting framework showcases promising efficacy, with the Catboost classifier, in conjunction with VAE, demonstrating notable accuracy compared to alternative classifiers. Rigorous evaluation employing diverse metrics—including accuracy, recall, F-score, precision, and receiver operating characteristic (ROC) curve analysis—attests to the robustness and generalization capacity of the proposed model.

In this study, we present an integrated approach utilizing IoT data and machine learning models to enhance precision agriculture. We collected an extensive IoT secondary dataset from an online data repository, including environmental parameters such as temperature, humidity, and soil nutrient levels, from various sensors deployed in agricultural fields. This dataset, consisting of over 1 million data points, provided comprehensive insights into the environmental conditions affecting crop yield. The data were preprocessed and used to develop predictive models for crop yield and recommendations. Our evaluation shows that the LightGBM, Decision Tree, and Random Forest classifiers achieved high accuracy scores of 98.90%, 98.48%, and 99.31%, respectively. The IoT data collection enabled real-time monitoring and accurate data input, significantly improving the models’ performance. These findings demonstrate the potential of combining IoT and machine learning to optimize resource use and improve crop management in smart farming. Future work will focus on expanding the dataset to include more diverse environmental factors and exploring the integration of advanced deep learning techniques for even more accurate predictions.

The widespread adoption and continuous development of social media platforms have greatly enhanced the availability of information, thanks to recent technical progress. Regarding these alterations, the question arises as to whether social media assessments have any impact on market price fluctuations, especially considering the volatility of pricing values and the complexity of the related area. The purpose of this research is to create prediction models that anticipate price fluctuations in the subsequent day by utilizing Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). We have developed a preprocessing pipeline that incorporates a range of natural language processing (NLP) tasks, including sentiment analysis, aggregation, and imputation of missing data. The purpose of this pipeline is to extract social and emotional data from the databases of prominent organizations, including Tesla and Amazon. We utilized LSTM and GRU models to train on different combinations of the novel features. The models demonstrate a robust correlation between stock prices and social media responses. More specifically, the GRU model attained a R2 score of 96