Borderline SMOTE Research Articles

X-ray Image Analysis for Dental Disease: A Deep Learning Approach Using EfficientNets

Dental cavities are a highly common persistent dental problem that impacts populations across different age groups on a global scale. It is crucial to get a dental issue diagnosed as early as possible and with as much accuracy as possible to treat it efficiently and prevent any related issues. If a dental infection is not treated, it will eventually grow and cause tooth loss. Dental X-ray images are crucial and beneficial in the diagnostic process of dental diseases for dentists. By applying Deep Learning (DL) techniques to dental X-ray images, dental experts can efficiently and precisely detect dental conditions, including dental cavities, fillings and implants. The objective of this research is to assess the performance of DL-based methods for dental disease detection via panoramic radiographs. In this study, we evaluated the performance of all of the EfficientNet variants (e.g., EfficientNets B0-B7) to determine which one is the most effective model for detecting dental disease. Moreover, we utilized the Borderline Synthetic Minority Oversampling Technique (SMOTE) to cope with the issue related to the minority classes contained in the dataset. To assess the efficacy of the model, various metrics are employed, including recall, accuracy, precision, loss, and F1-score. As a result, the performance of the EfficientNet-B5 model was superior to that of the other EfficientNet models. The EfficientNet-B5 model achieved the following values for its metrics: F1-score, accuracy, recall, AUC, and precision: 98.37%, 98.32%, 98.32%, 99.21%, and 98.32%, respectively. The accuracy rates for the EfficientNet-B0, EfficientNet-B1, EfficientNet-B2, EfficientNet-B3, EfficientNet-B4, EfficientNet-B6, and EfficientNet-B7, are 91.59%, 94.12%, 93.28%, 85.71%, 94.96%, 96.64% and 90.76%, respectively. The results indicated that the EfficientNet-B5 model performs better than other EfficientNet classifiers, which supports dental professionals significantly in the recognition of dental diseases.

A unified Foot and Mouth Disease dataset for Uganda: evaluating machine learning predictive performance degradation under varying distributions.

In Uganda, the absence of a unified dataset for constructing machine learning models to predict Foot and Mouth Disease outbreaks hinders preparedness. Although machine learning models exhibit excellent predictive performance for Foot and Mouth Disease outbreaks under stationary conditions, they are susceptible to performance degradation in non-stationary environments. Rainfall and temperature are key factors influencing these outbreaks, and their variability due to climate change can significantly impact predictive performance. This study created a unified Foot and Mouth Disease dataset by integrating disparate sources and pre-processing data using mean imputation, duplicate removal, visualization, and merging techniques. To evaluate performance degradation, seven machine learning models were trained and assessed using metrics including accuracy, area under the receiver operating characteristic curve, recall, precision and F1-score. The dataset showed a significant class imbalance with more non-outbreaks than outbreaks, requiring data augmentation methods. Variability in rainfall and temperature impacted predictive performance, causing notable degradation. Random Forest with borderline SMOTE was the top-performing model in a stationary environment, achieving 92% accuracy, 0.97 area under the receiver operating characteristic curve, 0.94 recall, 0.90 precision, and 0.92 F1-score. However, under varying distributions, all models exhibited significant performance degradation, with random forest accuracy dropping to 46%, area under the receiver operating characteristic curve to 0.58, recall to 0.03, precision to 0.24, and F1-score to 0.06. This study underscores the creation of a unified Foot and Mouth Disease dataset for Uganda and reveals significant performance degradation in seven machine learning models under varying distributions. These findings highlight the need for new methods to address the impact of distribution variability on predictive performance.

Predictive Modeling of COVID-19 Readmissions: Insights from Machine Learning and Deep Learning Approaches.

This project employs artificial intelligence, including machine learning and deep learning, to assess COVID-19 readmission risk in Malaysia. It offers tools to mitigate healthcare resource strain and enhance patient outcomes. This study outlines a methodology for classifying COVID-19 readmissions. It starts with dataset description and pre-processing, while the data balancing was computed through Random Oversampling, Borderline SMOTE, and Adaptive Synthetic Sampling. Nine machine learning and ten deep learning techniques are applied, with five-fold cross-validation for evaluation. Optuna is used for hyperparameter selection, while the consistency in training hyperparameters is maintained. Evaluation metrics encompass accuracy, AUC, and training/inference times. Results were based on stratified five-fold cross-validation and different data-balancing methods. Notably, CatBoost consistently excelled in accuracy and AUC across all tables. Using ROS, CatBoost achieved the highest accuracy (0.9882 ± 0.0020) with an AUC of 1.0000 ± 0.0000. CatBoost maintained its superiority in BSMOTE and ADASYN as well. Deep learning approaches performed well, with SAINT leading in ROS and TabNet leading in BSMOTE and ADASYN. Decision Tree ensembles like Random Forest and XGBoost consistently showed strong performance.

Infusing Weighted Average Ensemble Diversity for Advanced Breast Cancer Detection

ABSTRACTBreast cancer is a widespread health threat for women globally, often difficult to detect early due to its asymptomatic nature. As the disease advances, treatment becomes intricate and costly, ultimately resulting in elevated fatality rates. Currently, despite the widespread use of advanced machine learning (ML) and deep learning (DL) techniques, a comprehensive diagnosis of breast cancer remains elusive. Most of the existing methods primarily utilize either attention‐based deep models or models based on handcrafted features to capture and gather local details. However, both of these approaches lack the capability to offer essential local information for precise tumor detection. Additionally, the available breast cancer datasets suffer from class imbalance issue. Hence, this paper presents a novel weighted average ensemble network (WA‐ENet) designed for early‐stage breast cancer detection that leverages the ability of ensemble technique over single classifier‐based models for more robust and accurate prediction. The proposed model employs a weighted average‐based ensemble technique, combining predictions from three diverse classifiers. The optimal combination of weights is determined using the hill climbing (HC) algorithm. Moreover, the proposed model enhances overall system performance by integrating deep features and handcrafted features through the use of HOG, thereby providing precise local information. Additionally, the proposed work addresses class imbalance by incorporating borderline synthetic minority over‐sampling technique (BSMOTE). It achieves 99.65% accuracy on BUSI and 97.48% on UDIAT datasets.

Data oversampling and imbalanced datasets: an investigation of performance for machine learning and feature engineering

The classification of imbalanced datasets is a prominent task in text mining and machine learning. The number of samples in each class is not uniformly distributed; one class contains a large number of samples while the other has a small number. Overfitting of the model occurs as a result of imbalanced datasets, resulting in poor performance. In this study, we compare different oversampling techniques like synthetic minority oversampling technique (SMOTE), support vector machine SMOTE (SVM-SMOTE), Border-line SMOTE, K-means SMOTE, and adaptive synthetic (ADASYN) oversampling to address the issue of imbalanced datasets and enhance the performance of machine learning models. Preprocessing significantly enhances the quality of input data by reducing noise, redundant data, and unnecessary data. This enables the machines to identify crucial patterns that facilitate the extraction of significant and pertinent information from the preprocessed data. This study preprocesses the data using various top-level preprocessing steps. Furthermore, two imbalanced Twitter datasets are used to compare the performance of oversampling techniques with six machine learning models including random forest (RF), SVM, K-nearest neighbor (KNN), AdaBoost (ADA), logistic regression (LR), and decision tree (DT). In addition, the bag of words (BoW) and term frequency and inverse document frequency (TF-IDF) features extraction approaches are used to extract features from the tweets. The experiments indicate that SMOTE and ADASYN perform much better than other techniques thus providing higher accuracy. Additionally, overall results show that SVM with ’linear’ kernel tends to attain the highest accuracy and recall score of 99.67% and 1.00% on ADASYN oversampled datasets and 99.57% accuracy on SMOTE oversampled dataset with TF-IDF features. The SVM model using 10-fold cross-validation experiments achieved 97.40 mean accuracy with a 0.008 standard deviation. Our approach achieved 2.62% greater accuracy as compared to other current methods.

Financial risk forewarning with an interpretable ensemble learning approach: An empirical analysis based on Chinese listed companies

An efficient and intelligent forewarning model for financial risk is essential to assist company managers, investors, and market regulators in risk management. This research aims to establish an accurate and understandable financial distress forewarning model based on advanced ensemble learning techniques, including Bootstrap Aggregating (Bagging) and Light Gradient Boosting Machine (LightGBM). Meanwhile, the Borderline-Synthetic Minority Over-sampling Technique (Borderline-SMOTE) is employed to address the imbalance problem between financial distress and non-distressed samples. The Multi-Objective Particle Swarm Optimization (MOPSO) is applied to perform multi-objective optimization of the hyperparameters of the proposed model. In addition, the SHapley Additive exPlanations (SHAP) is adopted to conduct an interpretability analysis of the forewarning results. The Chinese listed companies from 2006 to 2022 are considered as the research samples, and the financial ratios of the sample companies in the first, second, and third years before the financial distress is utilized as forewarning features. The experimental findings reveal that the proposed model has the best forewarning performance by using the features from first year before the financial distress. The model proposed in this study also outperforms other comparative models in terms of Accuracy (88.45%), Negative Predictive Value (74.44%), Recall (91.10%), and Specificity (80.26%). Finally, this research has discovered that the Return on Total Assets is the most influential indicator in forewarning financial distress. Therefore, the proposed model can provide listed company managers, investors, and market regulators with an intelligent and effective tool in decision-making for financial distress forewarning.

Anomaly detection in wind turbine blades based on PCA and convolutional kernel transform models: employing multivariate SCADA time series analysis

With the widespread application of wind power technology, the detection of abnormalities in wind turbine blades has become a key research area. The use of data from monitoring and data acquisition (SCADA) systems for data-driven fault detection research presents new challenges. This study utilizes short-term SCADA data from wind turbine generators to classify the blade abnormal and normal operational states, thereby introducing a new method called PCABSMMR. This strategy integrates principal component analysis (PCA) and borderline-synthetic minority over-sampling technique (Borderline-SMOTE) for data processing and utilizes an improved multi-dimensional time series classification (MTSC) model. It combines one-dimensional convolution from deep learning with shallow learning’s rigid classifiers. PCA is used for dimensionality reduction, while Borderline-SMOTE expands the samples of minority class fault instances. Comparative analysis with various methods shows that the proposed method has an average F1-score of 0.98, outperforming many state-of-the-art MTSC models across various evaluation metrics.

An imbalance data quality monitoring based on SMOTE-XGBOOST supported by edge computing

Product assembly involves extensive production data that is characterized by high dimensionality, multiple samples, and data imbalance. The article proposes an edge computing-based framework for monitoring product assembly quality in industrial Internet of Things. Edge computing technology relieves the pressure of aggregating enormous amounts of data to cloud center for processing. To address the problem of data imbalance, we compared five sampling methods: Borderline SMOTE, Random Downsampling, Random Upsampling, SMOTE, and ADASYN. Finally, the quality monitoring model SMOTE-XGBoost is proposed, and the hyperparameters of the model are optimized by using the Grid Search method. The proposed framework and quality control methodology were applied to an assembly line of IGBT modules for the traction system, and the validity of the model was experimentally verified.

Synthetic minority over-sampling technique-enhanced machine learning models for predicting recurrence of postoperative chronic subdural hematoma.

Chronic subdural hematoma (CSDH) is a neurological condition with high recurrence rates, primarily observed in the elderly population. Although several risk factors have been identified, predicting CSDH recurrence remains a challenge. Given the potential of machine learning (ML) to extract meaningful insights from complex data sets, our study aims to develop and validate ML models capable of accurately predicting postoperative CSDH recurrence. Data from 447 CSDH patients treated with consecutive burr-hole irrigations at Wenzhou Medical University's First Affiliated Hospital (December 2014-April 2019) were studied. 312 patients formed the development cohort, while 135 comprised the test cohort. The Least Absolute Shrinkage and Selection Operator (LASSO) method was employed to select crucial features associated with recurrence. Eight machine learning algorithms were used to construct prediction models for hematoma recurrence, using demographic, laboratory, and radiological features. The Border-line Synthetic Minority Over-sampling Technique (SMOTE) was applied to address data imbalance, and Shapley Additive Explanation (SHAP) analysis was utilized to improve model visualization and interpretability. Model performance was assessed using metrics such as AUROC, sensitivity, specificity, F1 score, calibration plots, and decision curve analysis (DCA). Our optimized ML models exhibited prediction accuracies ranging from 61.0% to 86.2% for hematoma recurrence in the validation set. Notably, the Random Forest (RF) model surpassed other algorithms, achieving an accuracy of 86.2%. SHAP analysis confirmed these results, highlighting key clinical predictors for CSDH recurrence risk, including age, alanine aminotransferase level, fibrinogen level, thrombin time, and maximum hematoma diameter. The RF model yielded an accuracy of 92.6% with an AUC value of 0.834 in the test dataset. Our findings underscore the efficacy of machine learning algorithms, notably the integration of the RF model with SMOTE, in forecasting the recurrence of postoperative chronic subdural hematoma. Leveraging the RF model, we devised an online calculator that may serve as a pivotal instrument in tailoring therapeutic strategies and implementing timely preventive interventions for high-risk patients.

A comparative study in class imbalance mitigation when working with physiological signals.

Class imbalance is a common challenge that is often faced when dealing with classification tasks aiming to detect medical events that are particularly infrequent. Apnoea is an example of such events. This challenge can however be mitigated using class rebalancing algorithms. This work investigated 10 widely used data-level class imbalance mitigation methods aiming towards building a random forest (RF) model that attempts to detect apnoea events from photoplethysmography (PPG) signals acquired from the neck. Those methods are random undersampling (RandUS), random oversampling (RandOS), condensed nearest-neighbors (CNNUS), edited nearest-neighbors (ENNUS), Tomek's links (TomekUS), synthetic minority oversampling technique (SMOTE), Borderline-SMOTE (BLSMOTE), adaptive synthetic oversampling (ADASYN), SMOTE with TomekUS (SMOTETomek) and SMOTE with ENNUS (SMOTEENN). Feature-space transformation using PCA and KernelPCA was also examined as a potential way of providing better representations of the data for the class rebalancing methods to operate. This work showed that RandUS is the best option for improving the sensitivity score (up to 11%). However, it could hinder the overall accuracy due to the reduced amount of training data. On the other hand, augmenting the data with new artificial data points was shown to be a non-trivial task that needs further development, especially in the presence of subject dependencies, as was the case in this work.

Oversampling method via adaptive double weights and Gaussian kernel function for the transformation of unbalanced data in risk assessment of cardiovascular disease

In risk assessment of cardiovascular disease (CVD), the classification error caused by unbalanced data is a significant challenge, which has sparked widespread concern and research upsurge in the field of data mining. Therefore, in view of the imbalance of CVD data sets, an oversampling method via adaptive double weights and Gaussian kernel function (ADWGKFO) is proposed, which converts the unbalanced data sets into balanced data sets. Firstly, clustering algorithm is utilized to cluster minority samples, boundary samples are identified by Borderline-Synthetic Minority Over-sampling Technique (Borderline-SMOTE), K nearest neighbor and support vector machine algorithms, and the number of samples synthesized in each group is calculated based on the double weights of boundary points and majority distribution. Secondly, in order to clearly define the classification boundary, the mutual class potential of new samples in each cluster is calculated by Gaussian kernel function, and new samples are filtered according to the mutual class potential until the data set is balanced. Finally, taking the data sets from Kaggle platform as the research samples, the proposed method is empirically analyzed. In order to validate the efficacy and universality of the proposed method, this paper selects CatBoost that is a new integrated algorithm to test the effect of the ADWGKFO method, and compares it with different sampling methods and different classifiers using performance evaluation indexes such as accuracy, F1-score and area under the curve (AUC). Compared with the combinations of other methods, the accuracy, F1-score and AUC are significantly improved. It is concluded that the ADWGKFO method described in this paper can successfully improve the data quality, and increases the reliability of CVD risk assessment.

HBMD-Net: Feature Fusion Based Breast Cancer Classification with Class Imbalance Resolution.

Breast cancer, a widespread global disease, represents a significant threat to women's health and lives, ranking as one of the most vulnerable malignant tumors they face. Many researchers have proposed their computer-aided diagnosis systems for classifying breast cancer. The majority of these approaches primarily utilize deep learning (DL) methods, which are not entirely reliable. These approaches overlook the crucial necessity of incorporating both local and global information for precise tumor detection, despite the fact that the subtle nuances are crucial for precise breast cancer classification. In addition, there are a limited number of publicly available breast cancer datasets, and the ones that are available tend to be imbalanced in nature. Therefore, this paper presents the hybrid breast mass detection-network (HBMD-Net) to address two critical challenges: class imbalance and the need to recognize that relying solely on either global or local features falls short in achieving precise tumor classification. To overcome the problem of class imbalance, HBMD-Net incorporates the borderline synthetic minority over-sampling technique (BSMOTE). Simultaneously, it employs a feature fusion approach, combining features by utilizing ResNet50 to extract deep features that provide global information, while handcrafted features are derived using histogram orientation gradient (HOG), that provide local information. In addition, an ROI segmentation has been implemented to avoid misclassifications. This integrated strategy substantially enhances breast cancer classification performance. Moreover, the proposed method integrates the block matching and 3D (BM3D) denoising filter to effectively eliminate multiplicative noise that has enhanced the performance of the system. The evaluation of the proposed HBMD-Net encompasses two breast ultrasound (BUS) datasets, namely BUSI and UDIAT. The proposed model has demonstrated a satisfactory performance, achieving accuracies of 99.14% and 94.49% respectively.

INTERPRETABLE MACHINE LEARNING FOR PREDICTING RISK OF INVASIVE FUNGAL INFECTION IN CRITICALLY ILL PATIENTS IN THE INTENSIVE CARE UNIT: A RETROSPECTIVE COHORT STUDY BASED ON MIMIC-IV DATABASE.

The delayed diagnosis of invasive fungal infection (IFI) is highly correlated with poor prognosis in patients. Early identification of high-risk patients with invasive fungal infections and timely implementation of targeted measures is beneficial for patients. The objective of this study was to develop a machine learning-based predictive model for invasive fungal infection in patients during their intensive care unit (ICU) stay. Retrospective data was extracted from adult patients in the MIMIC-IV database who spent a minimum of 48 h in the ICU. Feature selection was performed using LASSO regression, and the dataset was balanced using the BL-SMOTE approach. Predictive models were built using six machine learning algorithms. The Shapley additive explanation algorithm was used to assess the impact of various clinical features in the optimal model, enhancing interpretability. The study included 26,346 ICU patients, of whom 379 (1.44%) were diagnosed with invasive fungal infection. The predictive model was developed using 20 risk factors, and the dataset was balanced using the borderline-SMOTE (BL-SMOTE) algorithm. The BL-SMOTE random forest model demonstrated the highest predictive performance (area under curve = 0.88, 95% CI = 0.84-0.91). Shapley additive explanation analysis revealed that the three most influential clinical features in the BL-SMOTE random forest model were dialysis treatment, APSIII scores, and liver disease. The machine learning model provides a reliable tool for predicting the occurrence of IFI in ICU patients. The BL-SMOTE random forest model, based on 20 risk factors, exhibited superior predictive performance and can assist clinicians in early assessment of IFI occurrence in ICU patients. Importance: Invasive fungal infections are characterized by high incidence and high mortality rates characteristics. In this study, we developed a clinical prediction model for invasive fungal infections in critically ill patients based on machine learning algorithms. The results show that the machine learning model based on 20 clinical features has good predictive value.

Investigating the impact of influential factors on crash types for autonomous vehicles at intersections

Autonomous Vehicles (AVs) are being promoted as an emerging technology with the potential to improve traffic efficiency and safety. However, the scarcity of publicly available AV crash data contributes to a limited understanding of safety issues for deploying AVs. To gain a deeper understanding of the factors contributing to AV crashes, this study created a unique dataset by combining 445 intersection AV crashes from the California Department of Motor Vehicles with detailed roadway geometric and traffic characteristic data. To address the issue of imbalanced data, the Borderline synthetic minority oversampling technique (BL-SMOTE) was utilized in this study. A framework for analyzing AV crash types at intersections was also developed using the Categorical gradient boosting (Catboost) and Bayesian networks (BNs) models. The results show that inappropriate vehicle maneuvers, such as sudden lane changes or speed control, can increase the likelihood of an AV being involved in a sideswipe crash. Additionally, in scenarios where the AADT of the major road is lower, AVs are more prone to rear-end crashes when proceeding. These findings can provide valuable insights for creating safety management strategies in different situations and improve the development of innovative car warning technologies, such as rear-end collision warnings.

3D Mineral Prospectivity Mapping from 3D Geological Models Using Return–Risk Analysis and Machine Learning on Imbalance Data

Three-dimensional Mineral Prospectivity Mapping (3DMPM) is an innovative approach to mineral exploration that combines multiple geological data sources to create a three-dimensional (3D) model of a mineral deposit. It provides an accurate representation of the subsurface that can be used to identify areas with mineral potential. These 3D geological models are the typical data source for 3D prospective modeling. Geological data sets from multiple sources are used to construct 3D geological models. Since in practice there is a significant imbalance in the ratio of mineralized to non-mineralized classes, the classification results will be biased in favor of the more observed classes. Borderline-SMOTE (BLSMOTE) is an oversampling technique used to solve the problem of unbalanced datasets and works by generating synthetic data points along the boundary line between the minority and majority classes. This helps to create a more balanced dataset without introducing too much noise. Non-mineralized samples can be generated by randomly selecting non-mineralized locations, which means that uncertainties are generated. In this paper, we take the shallow-forming low-temperature hydrothermal deposit Guizhou Lannigou gold deposit as an example to extract the ore-controlling elements and establish a 3D geological model. A total of 50 training samples are generated using the sampling method described above, and 50 mineralization prospects are generated using Random Forests. A return–risk analysis was used to explore the uncertainties associated with synthetic positive samples and randomly selected negative samples, and to determine the final mineral potential values. Based on the evaluation metrics G-mean and F-value, the model using BLSMOTE outperforms the model without the synthetic algorithm and the models using SMOTE and KMeansSMOTE. The optimal model BLSMOTE18 has an AUC of 0.9288. The methodology also performs superiorly with different levels of class imbalance datasets. Excluding the predictions where the results highly overlap with known deposits, five target zones were circled for the targets using a P-A plot, all of which have obvious metallogenic geological features. Among them, Target1 and Target2 have good potential for mineralization, which is of great significance for future mineral exploration work.

Determining Resampling Ratios Using BSMOTE and SVM-SMOTE for Identifying Rare Attacks in Imbalanced Cybersecurity Data

Machine Learning is widely used in cybersecurity for detecting network intrusions. Though network attacks are increasing steadily, the percentage of such attacks to actual network traffic is significantly less. And here lies the problem in training Machine Learning models to enable them to detect and classify malicious attacks from routine traffic. The ratio of actual attacks to benign data is significantly high and as such forms highly imbalanced datasets. In this work, we address this issue using data resampling techniques. Though there are several oversampling and undersampling techniques available, how these oversampling and undersampling techniques are most effectively used is addressed in this paper. Two oversampling techniques, Borderline SMOTE and SVM-SMOTE, are used for oversampling minority data and random undersampling is used for undersampling majority data. Both the oversampling techniques use KNN after selecting a random minority sample point, hence the impact of varying KNN values on the performance of the oversampling technique is also analyzed. Random Forest is used for classification of the rare attacks. This work is done on a widely used cybersecurity dataset, UNSW-NB15, and the results show that 10% oversampling gives better results for both BMSOTE and SVM-SMOTE.

A clustered borderline synthetic minority over-sampling technique for balancing quick access recorder data

Most of the flight accident data have uneven distribution of categories. When the traditional classifier is applied to this data, it will pay less attention to the minority class data. Synthetic Minority Over-sampling Technique (SMOTE), and its improvements are well-known methods to address this imbalance problem at the data level. However, traditional algorithms still have the problems in blurring the boundary of positive and negative classes and changing the distribution of original data. In order to overcome these problems and accurately predict flight accidents, a new Clustered Biased Borderline SMOTE(CBB-SMOTE) is proposed for Quick Access Recorder (QAR) Go-Around data. It generates more obvious positive and negative class boundaries by using K-means for boundary minority class data and safety minority class data respectively, and maintains the original data distribution to the greatest extent through a biased oversampling method. Experiments were carried out on a group of QAR Go-Around data. The data set is balanced by CBB-SMOTE, SMOTE, Cluster-SMOTE algorithm respectively, and the random forest algorithm is used to predict the new data set. The experimental results show that CBB-SMOTE outperforms the SMOTE in terms of G-means value, Recall and AUC.

Optimizing sleep staging on multimodal time series: Leveraging borderline synthetic minority oversampling technique and supervised convolutional contrastive learning

Sleep is an important research area in nutritional medicine that plays a crucial role in human physical and mental health restoration. It can influence diet, metabolism, and hormone regulation, which can affect overall health and well-being. As an essential tool in the sleep study, the sleep stage classification provides a parsing of sleep architecture and a comprehensive understanding of sleep patterns to identify sleep disorders and facilitate the formulation of targeted sleep interventions. However, the class imbalance issue is typically salient in sleep datasets, which severely affects classification performances. To address this issue and to extract optimal multimodal features of EEG, EOG, and EMG that can improve the accuracy of sleep stage classification, a Borderline Synthetic Minority Oversampling Technique (B-SMOTE)-Based Supervised Convolutional Contrastive Learning (BST-SCCL) is proposed, which can avoid the risk of data mismatch between various sleep knowledge domains (varying health conditions and annotation rules) and strengthening learning characteristics of the N1 stage from the pair-wise segments comparison strategy. The lightweight residual network architecture with a novel truncated cross-entropy loss function is designed to accommodate multimodal time series and boost the training speed and performance stability. The proposed model has been validated on four well-known public sleep datasets (Sleep-EDF-20, Sleep-EDF-78, ISRUC-1, and ISRUC-3) and its superior performance (overall accuracy of 91.31–92.34%, MF1 of 88.21–90.08%, and Cohen’s Kappa coefficient k of 0.87–0.89) has further demonstrated its effectiveness. It shows the great potential of contrastive learning for cross-domain knowledge interaction in precision medicine.

Diabetes mellitus early warning and factor analysis using ensemble Bayesian networks with SMOTE-ENN and Boruta

Diabetes mellitus (DM) has become the third chronic non-infectious disease affecting patients after tumor, cardiovascular and cerebrovascular diseases, becoming one of the major public health issues worldwide. Detection of early warning risk factors for DM is key to the prevention of DM, which has been the focus of some previous studies. Therefore, from the perspective of residents' self-management and prevention, this study constructed Bayesian networks (BNs) combining feature screening and multiple resampling techniques for DM monitoring data with a class imbalance in Shanxi Province, China, to detect risk factors in chronic disease monitoring programs and predict the risk of DM. First, univariate analysis and Boruta feature selection algorithm were employed to conduct the preliminary screening of all included risk factors. Then, three resampling techniques, SMOTE, Borderline-SMOTE (BL-SMOTE) and SMOTE-ENN, were adopted to deal with data imbalance. Finally, BNs developed by three algorithms (Tabu, Hill-climbing and MMHC) were constructed using the processed data to find the warning factors that strongly correlate with DM. The results showed that the accuracy of DM classification is significantly improved by the BNs constructed by processed data. In particular, the BNs combined with the SMOTE-ENN resampling improved the most, and the BNs constructed by the Tabu algorithm obtained the best classification performance compared with the hill-climbing and MMHC algorithms. The best-performing joint Boruta-SMOTE-ENN-Tabu model showed that the risk factors of DM included family history, age, central obesity, hyperlipidemia, salt reduction, occupation, heart rate, and BMI.

On the Quality of Synthetic Generated Tabular Data

Class imbalance is a common issue while developing classification models. In order to tackle this problem, synthetic data have recently been developed to enhance the minority class. These artificially generated samples aim to bolster the representation of the minority class. However, evaluating the suitability of such generated data is crucial to ensure their alignment with the original data distribution. Utility measures come into play here to quantify how similar the distribution of the generated data is to the original one. For tabular data, there are various evaluation methods that assess different characteristics of the generated data. In this study, we collected utility measures and categorized them based on the type of analysis they performed. We then applied these measures to synthetic data generated from two well-known datasets, Adults Income, and Liar+. We also used five well-known generative models, Borderline SMOTE, DataSynthesizer, CTGAN, CopulaGAN, and REaLTabFormer, to generate the synthetic data and evaluated its quality using the utility measures. The measurements have proven to be informative, indicating that if one synthetic dataset is superior to another in terms of utility measures, it will be more effective as an augmentation for the minority class when performing classification tasks.