Adaptation art image style transfer by integrating CSDA-FD algorithm and OSDA-DS algorithm

The electrocardiogram (ECG), being one of the most extensively used signals to monitor cardiovascular diseases (CVDs), captures the heart’s arrhythmias. Patients with such pathology are often monitored for extended periods of time, requiring data storage, and a very time-consuming off-line search of anomalies. This is especially inefficient when indicative patterns in the biological signals are infrequent, requiring more analysis time of medical doctors, and entailing a difficult visual search task for the diagnosis.In this paper, we propose an automated deep learning pipeline based on reservoir computing (RC), followed by principal component analysis (PCA) and one-class support vector machine (OC-SVM), that can be used to perform on-line learning and real-time anomaly detection of pathological conditions and therefore to raise warnings. The learning step requires fixed computational complexity and memory that are affordable on a low-power microcontroller unit (MCU). During the learning phase, it uses a very limited amount of input data, e.g., 60 KB, which makes this work suitable for fast personalization when device restart is required. The detection accuracy has been evaluated on the publicly available MIT-BIH arrhythmia dataset, also modified to mimic temporal sliding of input tensors on the ECG streaming data. Best F1 score and accuracy are 91.5%, 95.4% with variance over the processed data of 0.001.

The prevalence of deepfake technology has led to increased risks to biometric security, social media integrity, and the useof audio and video content for disinformation. In response, studies have progressed to provide efficient detecting techniques.Specifically, this paper uses anomaly-based classifiers to give a comparative analysis of zero-shot learning-based deepfakedetection. We evaluate two classifiers: One-Class Support Vector Machine (OCSVM) and Isolation Forest (IF) with threedifferent feature settings: Gabor features, Latent features, and Fused features (a mix of Gabor and Latent features). Importantmeasures like F1 Score, Accuracy, Precision, and Recall are used to assess how well the classifiers perform. Our resultsprovide important insights and future directions into the relationship between feature types and classifier performance in thesetting of zero-shot learning.

Identity recognition encounters with several problems especially in feature extraction and pattern classification. Electrocardiogram (ECG) is a quasi-periodic signal which has highly discriminative characteristics in a population for subject recognition. The personal identity verification in a random population using kernel-based binary and one-class Support Vector Machines (SVMs) has been considered by other biometric traits, but has been so far left aside for analysis of ECG signals. This paper investigates the effect of different parameters of data set size, labeling data, configuration of training and testing data sets, feature extraction, different recording sessions, and random partition methods on accuracy and error rates of these SVM classifiers. The experiments were carried out with defining a number of scenarios on ECG data sets designed rely on feature extractors which were modeled based on an autocorrelation in conjunction with linear and nonlinear dimension reduction methods. The experimental results show that Kernel Principal Component Analysis has lower error rate in binary and one-class SVMs on random unknown ECG data sets. Moreover, one-class SVM can be robust recognition algorithm for ECG biometric verification if the sufficient number of biometric samples is available.

BackgroundNon-alcoholic fatty liver disease (NAFLD) is a common chronic liver condition that affects a quarter of the global adult population. To date, only a few NAFLD risk prediction models have been developed for Chinese older adults aged ≥ 60 years. This study presented the development of a risk prediction model for NAFLD in Chinese individuals aged ≥ 60 years and proposed personalised health interventions based on key risk factors to reduce NAFLD incidence among the population.MethodsA cross-sectional survey was carried out among 9,041 community residents in Shanghai. Three NAFLD risk prediction models (I, II, and III) were constructed using multivariate logistic regression analysis based on the least absolute shrinkage and selection operator regression analysis, and random forest model to select individual characteristics, respectively. To determine the optimal model, the three models’ discrimination, calibration, clinical application, and prediction capability were evaluated using the receiver operating characteristic (ROC) curve, calibration plot, decision curve analysis, and net reclassification index (NRI), respectively. To evaluate the optimal model’s effectiveness, the previously published NAFLD risk prediction models (Hepatic steatosis index [HSI] and ZJU index) were evaluated using the following five indicators: accuracy, precision, recall, F1-score, and balanced accuracy. A dynamic nomogram was constructed for the optimal model, and a Bayesian network model for predicting NAFLD risk in older adults was visually displayed using Netica software.ResultsThe area under the ROC curve of Models I, II, and III in the training dataset was 0.810, 0.826, and 0.825, respectively, and that of the testing data was 0.777, 0.797, and 0.790, respectively. No significant difference was found in the accuracy or NRI between the models; therefore, Model III with the fewest variables was determined as the optimal model. Compared with the HSI and ZJU index, Model III had the highest accuracy (0.716), precision (0.808), recall (0.605), F1 score (0.692), and balanced accuracy (0.723). The risk threshold for Model III was 20%–80%. Model III included body mass index, alanine aminotransferase level, triglyceride level, and lymphocyte count.ConclusionsA dynamic nomogram and Bayesian network model were developed to identify NAFLD risk in older Chinese adults, providing personalized health management strategies and reducing NAFLD incidence.

Background Age-related macular degeneration (AMD), diabetic retinopathy (DR), drusen, choroidal neovascularization (CNV), and diabetic macular edema (DME) are significant causes of visual impairment globally. Optical coherence tomography (OCT) imaging has emerged as a valuable diagnostic tool for these ocular conditions. However, subjective interpretation and inter-observer variability highlight the need for standardized diagnostic approaches. Methods This study aimed to develop a robust deep learning model using artificial intelligence (AI) techniques for the automated detection of drusen, CNV, and DME in OCT images. A diverse dataset of 1,528 OCT images from Kaggle.com was used for model training. The performance metrics, including precision, recall, sensitivity, specificity, F1 score, and overall accuracy, were assessed to evaluate the model's effectiveness. Results The developed model achieved high precision (0.99), recall (0.962), sensitivity (0.985), specificity (0.987), F1 score (0.971), and overall accuracy (0.987) in classifying diseased and healthy OCT images. These results demonstrate the efficacy and efficiency of the model in distinguishing between retinal pathologies. Conclusion The study concludes that the developed deep learning model using AI techniques is highly effective in the automated detection of drusen, CNV, and DME in OCT images. Further validation studies and research efforts are necessary to evaluate the generalizability and integration of the model into clinical practice. Collaboration between clinicians, policymakers, and researchers is essential for advancing diagnostic tools and management strategies for AMD and DR. Integrating this technology into clinical workflows can positively impact patient care, particularly in settings with limited access to ophthalmologists. Future research should focus on collecting independent datasets, addressing potential biases, and assessing real-world effectiveness. Overall, the use of machine learning algorithms in conjunction with OCT imaging holds great potential for improving the detection and management of drusen, CNV, and DME, leading to enhanced patient outcomes and vision preservation.

Diabetes mellitus (DM) is a chronic disorder, characterized by impaired glucose metabolism. It is linked to increased risks of several diseases such as atrial fibrillation, cancer, and cardiovascular diseases. Therefore, DM prevention is essential. However, the traditional regression-based DM-onset prediction methods are incapable of investigating future DM for generally healthy individuals without DM. Employing gradient-boosting decision trees, we developed a machine learning-based prediction model to identify the DM signatures, prior to the onset of DM. We employed the nationwide annual specific health checkup records, collected during the years 2008 to 2018, from Kanazawa city, Ishikawa, Japan. The data included the physical examinations, blood and urine tests, and participant questionnaires. Individuals without DM (at baseline), who underwent more than two annual health checkups during the said period, were included. The new cases of DM onset were recorded when the participants were diagnosed with DM in the annual check-ups. The dataset was divided into three subsets in a 6:2:2 ratio to constitute the training, tuning (internal validation), and testing datasets. Employing the testing dataset, the ability of our trained prediction model to calculate the area under the curve (AUC), precision, recall, F1 score, and overall accuracy was evaluated. Using a 1,000-iteration bootstrap method, every performance test resulted in a two-sided 95% confidence interval (CI). We included 509,153 annual health checkup records of 139,225 participants. Among them, 65,505 participants without DM were included, which constituted36,303 participants in the training dataset and 13,101 participants in each of the tuning and testing datasets. We identified a total of 4,696 new DM-onset patients (7.2%) in the study period. Our trained model predicted the future incidence of DM with the AUC, precision, recall, F1 score, and overall accuracy of 0.71 (0.69-0.72 with 95% CI), 75.3% (71.6-78.8), 42.2% (39.3-45.2), 54.1% (51.2-56.7), and 94.9% (94.5-95.2), respectively. In conclusion, the machine learning-based prediction model satisfactorily identified the DM onset prior to the actual incidence.

Research on work-stress recognition for deep ground miners based on depth-separable convolutional neural network

IntroductionFault diagnosis analysis of mechanical equipment is greatly significant for maintaining the production efficiency of enterprises. Traditional diagnostic methods have shortcomings in accuracy and robustness.MethodsTherefore, the study integrates variational autoencoders with long short-term memory network models, enhances them using dropout methods, and proposes a hybrid diagnostic analysis model that combines improved autoencoder algorithms and signal reconstruction.ResultsThe experiment outcomes indicated that under the slow degradation mode of the bearing, the precision, recall, F1 score, and overall accuracy of the improved autoencoder model were 0.931, 0.933, 0.920, and 0.939, respectively, which were better than the pre-modified model. The fault diagnosis results showed that in the rapid degradation mode of the bearing, the research model discovered potential faults at 8,830 s, earlier than other models. The ablation experiment results showed that the precision, recall, F1 score, and overall accuracy of the enhanced study model using the dropout method were 0.83, 0.80, 0.82, and 0.99, respectively. Compared with the baseline model, the four indicators improved by 5.1%, 6.7%, 6.5%, and 5.3%, respectively. The memory usage test findings denoted that the average memory usage of the research model was less than 46%, which was better than the control model.DiscussionThe research promotes innovation and optimization of mechanical fault diagnosis technology, improves the accuracy and timeliness of fault diagnosis analysis models, and is of great significance for ensuring production safety, reducing maintenance costs, and improving enterprise economic benefits.

Machine learning (ML) models are increasingly being applied to diagnose and predict disease, but face technical challenges such as population drift, where the training and real-world deployed data distributions differ. This phenomenon can degrade model performance, risking incorrect diagnoses. Current detection methods are limited: not directly measuring population drift and often requiring ground truth labels for new patient data. Here, we propose using a one-class support vector machine (OCSVM) to detect population drift. We trained a OCSVM on the Wisconsin Breast Cancer dataset and tested its ability to detect population drift on simulated data. Simulated data was offset at 0.4 standard deviations of the minimum and maximum values of the radius_mean variable, at three noise levels: 5%, 10% and 30% of the standard deviation; 10,000 records per noise level. We hypothesised that increased noise would correlate with more OCSVM-detected inliers, indicating a sensitivity to population drift. As noise increased, more inliers were detected: 5% (27 inliers), 10% (486), and 30% (851). Therefore, this approach could effectively alert to population drift, supporting safe ML diagnostics adoption. Future research should explore OCSVM monitoring on real-world data, enhance model transparency, investigate complementary statistical and ML methods, and extend applications to other data types.

Anomaly detection using One-Class Support Vector Machine (OCSVM) have attracted wide attention in practical applications. Recent research focuses on enhancing OCSVM using either ensemble learning techniques or Multiple Kernel Learning (MKL) since single kernels such as the Gaussian Radial-Based Function (GRBF) kernel might not be flexible enough to construct a proper feature space. In this paper, we develop a new kernel, called centralized GRBF. Further, the two GRBF and centralized GRBF are combined by using a new ensemble kernel technique, called Coupled Ensemble-Kernels (CEK), to improve OCSVM for anomaly detection. Therefore, the final classification model is itself a large-margin classifier while it is actually an ensemble classifier coined with two sub-large-margin models. We show that the proposed CEK outperforms previous approaches using traditional ensemble learning methods and MKL for anomaly detection.KeywordsAnomaly DetectionKernel-Based MethodsOne-Class Support Vector MachineMultiple Kernel LearningEnsemble Learning

Accurate and up-to-date maps of built-up areas are crucial to support sustainable urban development. Earth Observation (EO) is a valuable data source to cover this demand. In particular, Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 MultiSpectral Instrument (MSI) missions offer new opportunities to map built-up areas on a global scale. Using Sentinel-2 images, recent urban mapping efforts achieved promising results by training Convolutional Neural Networks (CNNs) on available built-up data. However, these results strongly depend on the availability of local reference data for fully supervised training or assume that the application of CNNs to unseen areas (i.e. across-region generalization) produces satisfactory results. To alleviate these shortcomings, it is desirable to leverage Semi-Supervised Learning (SSL) algorithms that can take advantage of unlabeled data, especially because satellite data is plentiful. In this paper, we propose a novel Domain Adaptation (DA) approach using SSL that jointly exploits Sentinel-1 SAR and Sentinel-2 MSI to improve across-region generalization for built-up area mapping. Specifically, two identical sub-networks are incorporated into the proposed model to perform built-up area segmentation from SAR and optical images separately. Assuming that consistent built-up area segmentation should be obtained across data modality, we design an unsupervised loss for unlabeled data that penalizes inconsistent segmentation from the two sub-networks. Therefore, we propose to use complementary data modalities as real-world perturbations for consistency regularization. For the final prediction, the model takes both data modalities into account. Experiments conducted on a test set comprised of sixty representative sites across the world showed that the proposed DA approach achieves strong improvements (F1 score 0.694) over fully supervised learning from Sentinel-1 SAR data (F1 score 0.574), Sentinel-2 MSI data (F1 score 0.580) and their input-level fusion (F1 score 0.651). To demonstrate the effectiveness of DA, we also performed a comparison with two state-of-the-art products, namely GHS-BUILT-S2 and WSF 2019, on the test set. The comparison showed that our model is capable of producing built-up area maps with comparable or even better quality than the state-of-the-art global human settlement maps. Therefore, the multi-modal DA offers great potential to be adapted to produce easily updateable human settlements maps at a global scale.

The food nutrients in daily life is incredible source to make good food for good life cycle. In recent analysis, the new food styles and ingredients affects the public health especially children to cause various disease and impacts which leads dangerous to healthier life. By analyzing food nutrients and presence healthy ingredients in daily life is important. World health organization pays number of research to healthcare management and suggestions to improve the public health by recommending different protocols. The risk is identifying presence of nutrients scaling is tedious due to prevailing techniques machine learning techniques does not provide good results to make better good food recommendation. The problem is more data labels and features are need to analyze which increase the false rate to reduce accuracy. To address the issue, to propose a deep LSTM gated recurrent neural network (LSTM-GRNN) based on Support vector feature selection (SVFS) to identify the presence of food nutrients to recommend the good food to improve the public health. Initially the food product and nutrients scaling logs are collected and to make preprocessing based on C-score normalization form feature labels. Then nutrients scaling impact rate is analyzed for each food ingredients presence by extracting the margins from features list. The important feature is selected by using SVSR to reduce the feature margins based on support vector. Then the selected features are trained into LSTM unit with recurrent neural network to identify the presence of nutrients rate in each food and to categorize the class by presence. The higher scaling rate of class is considered as good food to recommend for healthy facts. The proposed system attains higher detection accuracy in precision, recall rate and f1 measure, also the lower false negative rate increase the performance of accuracy compared to the existing system.

AdaBoost.M1 has been successfully applied to improve the accuracy of a learning algorithm for multi-class classification problems. However, it may be hard to satisfy the required conditions in some practical cases. An improved algorithm called AdaBoost.MK is developed to solve this problem. Early proposed support vector machines (SVM)-based, multi-class classification algorithms work by splitting the original problem into a set of two-class subproblems. The amount of time and space required by these algorithms is very demanding. We develop a multi-class classification algorithm by incorporating one-class SVMs with a well-designed discriminant function. Finally, a hybrid method integrating AdaBoost.MK and one-class SVMs is proposed to solve multi-class classification problems. Experimental results on data sets from UCI and Statlog show that the proposed approach outperforms other multi-class algorithms, such as support vector data descriptions (SVDDs) and AdaBoost.M1 with one-class SVMs, and the improvement is found to be statistically significant.

Detecting low birth weight is crucial for early identification of at-risk pregnancies which are associated with significant neonatal and maternal morbidity and mortality risks. This study presents an efficient and interpretable framework for unsupervised detection of low, very low, and extreme birth weights. While traditional approaches to managing class imbalance require labeled data, our study explores the use of unsupervised learning to detect anomalies indicative of low birth weight scenarios. This method is particularly valuable in contexts where labeled data are scarce or labels for the anomaly class are not available, allowing for preliminary insights and detection that can inform further data labeling and more focused supervised learning efforts. We employed fourteen different anomaly detection algorithms and evaluated their performance using Area Under the Receiver Operating Characteristics (AUCROC) and Area Under the Precision-Recall Curve (AUCPR) metrics. Our experiments demonstrated that One Class Support Vector Machine (OCSVM) and Empirical-Cumulative-distribution-based Outlier Detection (ECOD) effectively identified anomalies across different birth weight categories. The OCSVM attained an AUCROC of 0.72 and an AUCPR of 0.0253 for extreme LBW detection, while the ECOD model showed competitive performance with an AUCPR of 0.045 for very low LBW cases. Additionally, a novel feature perturbation technique was introduced to enhance the interpretability of the anomaly detection models by providing insights into the relative importance of various prenatal features. The proposed interpretation methodology is validated by the clinician experts and reveals promise for early intervention strategies and improved neonatal care.

In recent years, precision agriculture has been researched to increase crop production with less inputs, as a promising means to meet the growing demand of agriculture products. Computer vision-based crop detection with unmanned aerial vehicle (UAV)-acquired images is a critical tool for precision agriculture. However, object detection using deep learning algorithms rely on a significant amount of manually prelabeled training datasets as ground truths. Field object detection, such as bales, is especially difficult because of (1) long-period image acquisitions under different illumination conditions and seasons; (2) limited existing prelabeled data; and (3) few pretrained models and research as references. This work increases the bale detection accuracy based on limited data collection and labeling, by building an innovative algorithms pipeline. First, an object detection model is trained using 243 images captured with good illimitation conditions in fall from the crop lands. In addition, domain adaptation (DA), a kind of transfer learning, is applied for synthesizing the training data under diverse environmental conditions with automatic labels. Finally, the object detection model is optimized with the synthesized datasets. The case study shows the proposed method improves the bale detecting performance, including the recall, mean average precision (mAP), and F measure (F1 score), from averages of 0.59, 0.7, and 0.7 (the object detection) to averages of 0.93, 0.94, and 0.89 (the object detection + DA), respectively. This approach could be easily scaled to many other crop field objects and will significantly contribute to precision agriculture.