Imbalance Issue Research Articles

Advancing the accuracy of tyrosinase inhibitory peptides prediction via a multiview feature fusion strategy

Tyrosinase plays a crucial role as an enzyme in the production of melanin, which is the pigment accountable for determining the color of the hair, eyes, and skin. Tyrosinase inhibitory peptides (TIPs), mainly designed to regulate the activity of the enzyme tyrosinase, are of interest in various domains, including cosmetics, dermatology, and pharmaceuticals, due to their potential applications in controlling skin pigmentation. To date, a few machine learning-based models have been proposed for predicting TIPs, but their predictive performance remains unsatisfactory. In this study, we propose an innovative computational approach, named TIPred-MVFF, to accurately predict TIPs using only sequence information. Firstly, we established an up-to-date and high-quality dataset by collecting samples from various sources. Secondly, we applied a multi-view feature fusion (MVFF) strategy to extract and explore probability and category information embedded in TIPs, employing several machine learning (ML) algorithms coupled with different commonly used sequence-based feature encodings. Then, we employed resampling approaches to address the class imbalance issue. Finally, to maximize the utility of each feature, we fused probability-based and sequence-based features, generating more informative feature that were used to develop the final prediction model. Based on the independent test, experimental results showed that TIPred-MVFF outperformed several conventional ML classifiers and existing methods in terms of prediction accuracy and robustness, achieving an accuracy of 0.937 and a Matthew’s correlation coefficient of 0.847. This new computational approach is anticipated to aid community-wide efforts in rapidly and cost-effectively discovering novel peptides with strong tyrosinase inhibitory activities.

A Novel Ensemble Belief Rule-Based Model for Online Payment Fraud Detection

In recent years, with the rapid development of technology and the economy, online transaction fraud has become more and more frequent. In the face of massive records of online transaction data, manual detection methods are long outdated, and machine learning methods have become mainstream. However, although traditional machine learning methods perform well in fraud detection tasks, the lack of interpretability and class imbalance issues have always been pain points that are difficult to resolve for such methods. Unlike traditional methods, the belief rule base, as a rule-based expert system model, can integrate expert knowledge and has excellent interpretability. In this paper, we propose an innovative ensemble BRB (belief rule base) model to solve the credit card fraud detection problem by combining an ensemble learning framework with the BRB model. Compared with traditional machine learning methods, the proposed model has the advantage of high interpretability. And compared with traditional BRB models, the ensemble framework enables better performance in dealing with highly imbalanced classification tasks. In an experimental study, two datasets of credit card fraud detection from Kaggle are used to validate the effectiveness of this work. The results show that this new method can achieve excellent performance in the application of fraud detection and is capable of effectively mitigating the impact of an imbalanced dataset.

A Review of Deep Learning Applications in Intrusion Detection Systems: Overcoming Challenges in Spatiotemporal Feature Extraction and Data Imbalance

In the rapid development of the Internet of Things (IoT) and large-scale distributed networks, Intrusion Detection Systems (IDS) face significant challenges in handling complex spatiotemporal features and addressing data imbalance issues. This article systematically reviews recent advancements in applying deep learning techniques in IDS, focusing on the core challenges of spatiotemporal feature extraction and data imbalance. First, this article analyzes the spatiotemporal dependencies of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) in network traffic feature extraction and examines the main methods these models use to solve this problem. Next, the impact of data imbalance on IDS performance is explored, and the effectiveness of various data augmentation and handling techniques, including Generative Adversarial Networks (GANs) and resampling methods, in improving the detection of minority class attacks is assessed. Finally, the paper highlights the current research gaps and proposes future research directions to optimize deep learning models further to enhance the detection capabilities and robustness of IDS in complex network environments. This review provides researchers with a comprehensive perspective, helping them identify the challenges in the current field and laying a foundation for future research efforts.

Machine learning models for quantitatively prediction of toxicity in macrophages induced by metal oxide nanoparticles.

As nanotechnology advances, metal oxide nanoparticles (MeONPs) increasingly come into contact with humans. The inhaled MeONPs cannot be effectively cleared by cilia or lung mucus. In the last decade, potential immune toxicity arising from exposure to MeONPs has been extensively debated, as lung macrophage is the main pathway for cleaning inhaled exogenous particles. However, their toxicity on lung macrophages has rarely been quantitatively predicted in silico due to the complexity of responses in macrophages and the intricate properties of MeONPs. Here, machine learning (ML) methods were used to establish models for quantitatively predicting the toxicity of MeONPs in macrophages. A multidimensional dataset including 240 data points covering the lethality, biochemical behaviors, and physicochemical properties of 30 MeONPs was obtained. ML models based on different algorithms with high prediction accuracy were constructed by addressing the issue of class imbalance during the training process. The models were verified by 10-fold cross-validation and external validation. The best-performed model has an R2 of 0.85 and 0.90 in the 10-fold cross-validation and external test set, respectively; and Q2 of 0.88 and 0.90 in the 10-fold cross-validation and test set, respectively. Five parameters that impact toxicity were identified and the toxicity mechanisms were elucidated by ML analysis. The prediction results can be used to fill the data gap in the risk assessment of nanomaterials. The framework offers valuable insights for designing and utilizing safe nanoparticles, as well as aiding in decision-making processes aimed at protecting the environment and public health.

Exploring Ensemble Learning Techniques for Infant Mortality Prediction: A Technical Analysis of XGBoost Stacking AdaBoost and Bagging Models.

Infant mortality remains a critical public health issue, reflecting the overall health and well-being of a population. Accurate prediction of infant mortality is crucial, as it enables healthcare providers to identify at-risk populations and implement targeted interventions. By analyzing factors such as maternal education, prenatal care access, nutrition, and environmental influences, predictions help in designing effective programs aimed at reducing infant deaths. This research paper aims to predict infant mortality in India by employing ensemble learning techniques, specifically eXtreme gradient boosting (XGBoost), stacking, adaptive boosting, and bagging. The data for the analysis are sourced from national surveys and demographic studies focusing on infant mortality in India. The collected data underwent rigorous preprocessing steps to prepare it for predictive modeling. Each ensemble learning model is applied to predict infant mortality rates based on the preprocessed data. The XGBoost handles complex and non-linear relationships within the data, and the stacking model is used for the accurate and robust predictions. The adaptive boosting model iteratively trains multiple weak learners, which makes the predictive model as stronger. The adaptive boosting technique enhances the performance of weak classifiers while effectively addressing class imbalance issues. Further, the bagging approach is implemented to derive the linear and non-linear relationships of infant mortality. Models were optimized using k-fold cross-validation to fine-tune their hyper parameters. The predictive ability of the ensemble techniques is analyzed by deploying using different performance parameters. XGBoost attained superior performance results, with a 98.75% accuracy, 98.56% precision, and 98.24% recall. The adaptive boosting model strengthened weak learners and addressed class imbalance issues, while the bagging method captures linear and non-linear relationships. Ensemble learning models demonstrated effectiveness in predicting infant mortality, with XGBoost excelling in handling complex and non-linear relationships. The simulation results revealed that ensemble learning models are highly effective in predicting infant mortality rates in India, with significant regional disparities observed. For example, the Northeast region exhibited the highest predicted infant mortality rates, while the South region recorded the lowest. These findings underscore the need for targeted interventions in high-mortality areas to reduce disparities. The study highlights the efficacy of ensemble learning models, particularly XGBoost, in predicting infant mortality in India. The findings emphasize the critical role of improving maternal education, access to prenatal care, and reducing socioeconomic disparities.

Object detection models and active learning for improvement of e-waste collection management systems in Korea.

This study identifies and analyzes issues within the management system of the waste home appliances free pickup service and seeks to enhance the system by using an object detection model. To overcome the limitations of manually inspecting approximately 5,000 collections per day, the YOLOv8 model was implemented. Photos for proof of collection, which were difficult to verify visually, were excluded from the image data. Labeling was performed on items defined by waste throughput, resulting in a dataset of 19,101 images. The initial training model achieved performance metrics of 0.950 mAP50 and 0.888 mAP50-95. The detection process for 11,003 images, including saving a summary file, took 7min and 32.8s. This method allows the system to automatically identify discrepancies between collection managers' registered data and the actual items collected. Additionally, active learning methods are proposed as a future enhancement strategy for the model. To improve future model performance, some data samples could be selected based on uncertainty by assigning weights to address class imbalance. The model generates bounding boxes for its predictions, and human annotators can verify these results, thereby reducing the cost of manual labeling. This approach can contribute to improving learning efficiency for any new data added later. Experimental results demonstrate that this method effectively resolves class imbalance issues and improves model performance through uncertainty sampling. It is expected that this approach will improve the existing manual inspection systems and maximize the efficiency of the future management process.

A novel oversampling method based on Wasserstein CGAN for imbalanced classification

Class imbalance is a crucial challenge in classification tasks, and in recent years, with the advancements in deep learning, research on oversampling techniques based on GANs has proliferated. These techniques have proven to be excellent in addressing the class imbalance issue by capturing the distributional features of minority samples during training and generating high-quality new samples. However, oversampling methods based on GANs may suffer from gradient vanishing, resulting in mode collapse, and produce noise and boundary-blurring issues when generating new samples. This paper proposes a novel oversampling method based on a conditional GAN (CGAN) incorporating Wasserstein distance. It generates an initial balanced dataset from minority class samples using the CGAN oversampling approach and then uses a noise and boundary recognition method based on K-means and k nearest neighbors algorithm to address the noise and boundary-blurring issues. The proposed method generates new samples that are highly consistent with the original sample distribution and effectively solves the problems of noise data and class boundary blurring. Experimental results on multiple public datasets show that the proposed method achieves significant improvements in evaluation metrics such as Recall, F1_score, G-mean, and AUC.

A deep learning based model for diabetic retinopathy grading

Diabetic retinopathy stands as a leading cause of blindness among people. Manual examination of DR images is labor-intensive and prone to error. Existing methods to detect this disease often rely on handcrafted features which limit the adaptability and classification accuracy. Thus, the aim of this research is to develop an automated and efficient system for early detection and accurate grading of diabetic retinopathy severity with less time consumption. In our research, we have developed a deep neural network named RSG-Net (Retinopathy Severity Grading) to classify DR into 4 stages (multi-class classification) and 2 stages (binary classification). The dataset utilized in this study is Messidor-1. In preprocessing, we have used Histogram Equalization to improve image contrast and denoising techniques to remove noise and artifacts which enhanced the clarity of the fundus images. We applied data augmentation techniques to our preprocessed images in order to tackle class imbalance issues. Augmentation techniques involve flipping, rotation, zooming and adjustment of color, contrast and brightness. The proposed RSG-Net model contains convolutional layers to perform automatic feature extraction from the input images and batch normalization layers to improve training speed and performance. The model also contains max pooling, drop out and fully connected layers. Our proposed RSG-Net model achieved a testing accuracy of 99.36%, specificity of 99.79% and a sensitivity of 99.41% in classifying diabetic retinopathy into 4 grades and it achieved 99.37% accuracy, 100% sensitivity and 98.62% specificity in classifying DR into 2 grades. The performance of RSG-Net is also compared with other state-of-the-art methodologies where it outperformed these methods.

LCSC-UAVNet: A High-Precision and Lightweight Model for Small-Object Identification and Detection in Maritime UAV Perspective

Unmanned Aerial Vehicle (UAV) object detection is crucial in various fields, such as maritime rescue and disaster investigation. However, due to small objects and the limitations of UAVs’ hardware and computing power, detection accuracy and computational overhead are the bottleneck issues of UAV object detection. To address these issues, a novel convolutional neural network (CNN) model, LCSC-UAVNet, is proposed, which substantially enhances the detection accuracy and saves computing resources. To address the issues of low parameter utilization and insufficient detail capture, we designed the Lightweight Shared Difference Convolution Detection Head (LSDCH). It combines shared convolution layers with various differential convolution to enhance the detail capture ability for small objects. Secondly, a lightweight CScConv module was designed and integrated to enhance detection speed while reducing the number of parameters and computational cost. Additionally, a lightweight Contextual Global Module (CGM) was designed to extract global contextual information from the sea surface and features of small objects in maritime environments, thus reducing the false negative rate for small objects. Lastly, we employed the WIoUv2 loss function to address the sample imbalance issue of the datasets, enhancing the detection capability. To evaluate the performance of the proposed algorithm, experiments were performed across three commonly used datasets: SeaDroneSee, AFO, and MOBdrone. Compared with the state-of-the-art algorithms, the proposed model showcases improvements in mAP, recall, efficiency, where the mAP increased by over 10%. Furthermore, it utilizes only 5.6 M parameters and 16.3 G floating-point operations, outperforming state-of-the-art models such as YOLOv10 and RT-DETR.

A Survey of Methods for Addressing Imbalance Data Problems in Agriculture Applications

This survey explores recent advances in addressing class imbalance issues for developing machine learning models in precision agriculture, with a focus on techniques used for plant disease detection, soil management, and crop classification. We examine the impact of class imbalance on agricultural data and evaluate various resampling methods, such as oversampling and undersampling, as well as algorithm-level approaches, to mitigate this challenge. The paper also highlights the importance of evaluation metrics, including F1-score, G-mean, and MCC, in assessing the performance of machine learning models under imbalanced conditions. Additionally, the review provides an in-depth analysis of emerging trends in the use of generative models, like GANs and VAEs, for data augmentation in agricultural applications. Despite the significant progress, challenges such as noisy data, incomplete datasets, and lack of publicly available datasets remain. This survey concludes with recommendations for future research directions, including the need for robust methods that can handle high-dimensional agricultural data effectively.

Lightweight GAN-Assisted Class Imbalance Mitigation for Apple Flower Bud Detection

Multi-class object detectors often suffer from the class imbalance issue, where substantial model performance discrepancies exist between classes. Generative adversarial networks (GANs), an emerging deep learning research topic, are able to learn from existing data distributions and generate similar synthetic data, which might serve as valid training data for improving object detectors. The current study investigated the utility of lightweight unconditional GAN in addressing weak object detector class performance by incorporating synthetic data into real data for model retraining, under an agricultural context. AriAplBud, a multi-growth stage aerial apple flower bud dataset was deployed in the study. A baseline YOLO11n detector was first developed based on training, validation, and test datasets derived from AriAplBud. Six FastGAN models were developed based on dedicated subsets of the same YOLO training and validation datasets for different apple flower bud growth stages. Positive sample rates and average instance number per image of synthetic data generated by each of the FastGAN models were investigated based on 1000 synthetic images and the baseline detector at various confidence thresholds. In total, 13 new YOLO11n detectors were retrained specifically for the two weak growth stages, tip and half-inch green, by including synthetic data in training datasets to increase total instance number to 1000, 2000, 4000, and 8000, respectively, pseudo-labeled by the baseline detector. FastGAN showed its resilience in successfully generating positive samples, despite apple flower bud instances being generally small and randomly distributed in the images. Positive sample rates of the synthetic datasets were negatively correlated with the detector confidence thresholds as expected, which ranged from 0 to 1. Higher overall positive sample rates were observed for the growth stages with higher detector performance. The synthetic images generally contained fewer detector-detectable instances per image than the corresponding real training images. The best achieved YOLO11n AP improvements in the retrained detectors for tip and half-inch green were 30.13% and 14.02% respectively, while the best achieved YOLO11n mAP improvement was 2.83%. However, the relationship between synthetic training instance quantity and detector class performances had yet to be determined. GAN was concluded to be beneficial in retraining object detectors and improving their performances. Further studies are still in need to investigate the influence of synthetic training data quantity and quality on retrained object detector performance.

Advanced sleep disorder detection using multi-layered ensemble learning and advanced data balancing techniques

Sleep disorder detection has greatly improved with the integration of machine learning, offering enhanced accuracy and effectiveness. However, the labor-intensive nature of diagnosis still presents challenges. To address these, we propose a novel coordination model aimed at improving detection accuracy and reliability through a multi-model ensemble approach. The proposed method employs a multi-layered ensemble model, starting with the careful selection of N models to capture essential features. Techniques such as thresholding, predictive scoring, and the conversion of Softmax labels into multidimensional feature vectors improve interpretability. Ensemble methods like voting and stacking are used to ensure collaborative decision-making across models. Both the original dataset and one modified using the Synthetic Minority Oversampling Technique (SMOTE) were evaluated to address data imbalance issues. The ensemble model demonstrated superior performance, achieving 96.88% accuracy on the SMOTE-implemented dataset and 95.75% accuracy on the original dataset. Moreover, an eight-fold cross-validation yielded an impressive 99.5% accuracy, indicating the reliability of the model in handling unbalanced data and ensuring precise detection of sleep disorders. Compared to individual models, the proposed ensemble method significantly outperformed traditional models. The combination of models not only enhanced accuracy but also improved the system's ability to handle unbalanced data, a common limitation in traditional methods. This study marks a significant advancement in sleep disorder detection through the integration of innovative ensemble techniques. The proposed approach, combining multiple models and advanced interpretability methods, promises improved patient outcomes and greater diagnostic accuracy, paving the way for future applications in medical diagnostics.

Enhanced ResNet-50 for garbage classification: Feature fusion and depth-separable convolutions.

As people's material living standards continue to improve, the types and quantities of household garbage they generate rapidly increase. Therefore, it is urgent to develop a reasonable and effective method for garbage classification. This is important for resource recycling and environmental improvement and contributes to the sustainable development of production and the economy. However, existing deep learning-based garbage image classification models generally suffer from low classification accuracy, insufficient robustness, and slow detection speed due to the large number of model parameters. To this end, a new garbage image classification model is proposed, with the ResNet-50 network as the core architecture. Specifically, first, a redundancy-weighted feature fusion module is proposed, enabling the model to fully leverage valuable feature information, thereby improving its performance. At the same time, the module filters out redundant information from multi-scale features, reducing the number of model parameters. Second, the standard 3×3 convolutions in ResNet-50 are replaced with depth-separable convolutions, significantly improving the model's computational efficiency while preserving the feature extraction capability of the original convolutional structure. Finally, to address the issue of class imbalance, a weighting factor is added to the Focal Loss, aiming to mitigate the negative impact of class imbalance on model performance and enhance the model's robustness. Experimental results on the TrashNet dataset show that the proposed model effectively reduces the number of parameters, improves detection speed, and achieves an accuracy of 94.13%, surpassing the vast majority of existing deep learning-based waste image classification models, demonstrating its solid practical value.

Machine Learning-Based Stacking Ensemble Model for Prediction of Heart Disease with Explainable AI and K-Fold Cross-Validation: A Symmetric Approach

One of the most complex and prevalent diseases is heart disease (HD). It is among the main causes of death around the globe. With changes in lifestyles and the environment, its prevalence is rising rapidly. The prediction of the disease in its early stages is crucial, as delays in diagnosis can cause serious complications and even death. Machine learning (ML) can be effective in this regard. Many researchers have used different techniques for the efficient detection of the disease and to overcome the drawbacks of existing models. Several ensemble models have also been applied. We proposed a stacking ensemble model named NCDG, which uses Naive Bayes, Categorical Boosting, and Decision Tree as base learners, with Gradient Boosting serving as the meta-learner classifier. We performed preprocessing using a factorization method to convert string columns into integers. We employ the Synthetic Minority Oversampling TEchnique (SMOTE) and BorderLineSMOTE balancing techniques to address the issue of data class imbalance. Additionally, we implemented hard and soft voting using voting classifier and compared the results with the proposed stacking model. For the Artificial Intelligence-based eXplainability of our proposed NCDG model, we use the SHapley Additive exPlanations (SHAP) technique. The outcomes show that our suggested stacking model, NCDG, performs better than the benchmark existing techniques. The experimental results of our proposed stacking model achieved the highest accuracy, F1-Score, precision and recall of 0.91, 0.91, 0.91 and 0.91, respectively, and an execution time of 653 s. Moreover, we have also utilized K-Fold Cross-Validation method to validate our predicted results. It is worth mentioning that our prediction results and their validation strongly coincide with each other which proves our approach to be symmetric.

Shiba: A versatile computational method for systematic identification of differential RNA splicing across platforms.

Alternative pre-mRNA splicing (AS) is a fundamental regulatory process that generates transcript diversity and cell type variation. We developed Shiba, a comprehensive method that integrates transcript assembly, splicing event identification, read counting, and differential splicing analysis across RNA-seq platforms. Shiba excels in capturing annotated and unannotated AS events with superior accuracy, sensitivity, and reproducibility. It addresses the often-overlooked issue of junction read imbalance, significantly reducing false positives to aid target prioritization and downstream analyses. Unlike other tools that require large numbers of biological replicates or resulting in low sensitivity and high false positives, Shiba's statistics framework is agnostic to sample size, as demonstrated by simulated data and its effective application to real n =1 RNA-seq datasets. To extend its utility to single-cell RNA-seq, we developed scShiba, which applies Shiba's pseudobulk approach to analyze splicing at the cluster level. scShiba successfully revealed AS regulation in developmental dopaminergic neurons and differences between excitatory and inhibitory neurons. Both Shiba and scShiba are available in Docker/Singularity containers and Snakemake pipelines, ensuring reproducibility. With their comprehensive capabilities, Shiba and scShiba enable systematic quantification of alternative splicing events across various platforms, laying a solid foundation for mechanistic exploration of the functional complexity in RNA splicing.

ASS-CD: Adapting Segment Anything Model and Swin-Transformer for Change Detection in Remote Sensing Images

Change detection (CD) is a critical task in analyzing the geographic information changes in remote sensing images (RSIs), yet it still faces challenges such as complex background interference, multi-scale varying objects, and class imbalance between positive and negative samples. Recently, with the development of pre-training and fine-tuning techniques, transferring the general knowledge embedded in large-scale pre-trained visual foundation models (PVFMs) to various downstream tasks has attracted significant attention. However, when directly applying these PVFMs to CD tasks in RSIs, the domain knowledge differences often result in unsatisfactory outcomes. To address the above issues, we propose a novel hierarchical adapter framework to efficiently adapt PVFMs like FastSAM and Swin-Transformer for CD task in RSIs, namely ASS-CD. The proposed method leverages lightweight adapter modules with a cross-attention mechanism, which not only preserves the general knowledge of PVFMs but also integrates global and local information, significantly enhancing CD accuracy. Further, the convolutional block attention module (CBAM) is adopted to reduce interference from complex backgrounds and focus on multi-scale objects, and the hierarchical deep supervision module (HDSM) is utilized to impose deep supervision on multi-scale feature maps and compute the Dice loss, addressing the issue of class imbalance in CD datasets. The experimental results on three widely used datasets demonstrate that our ASS-CD achieves the state-of-the-art performance, with an approximately 5% improvement on the LEVIR-CD dataset compared to the other CD methods.

A Trustworthy Framework for Skin Cancer Detection Using a CNN with a Modified Attention Mechanism

The early and accurate detection of skin cancer can reduce mortality rates and improve patient outcomes, but requires advanced diagnostics. The integration of artificial intelligence (AI) into healthcare enables the precise and timely detection of skin cancer. However, significant challenges remain including the difficulty in differentiating visually similar skin conditions and the limitations of diverse, representative datasets. In this study, we proposed DCAN-Net, a novel deep-learning framework designed for the early detection of skin cancer. The model leverages an efficient backbone architecture optimized for capturing diverse skin patterns, utilizing carefully tuned parameters to enhance the discrimination capabilities and refine the extracted features using modified attention modules, thereby prioritizing relevant foreground information while minimizing background noise. Furthermore, the Grad-CAM explainable AI method was employed, highlighting the most salient features within dermatoscopic images. The fused optimal feature representations significantly enhanced the dermatoscopic image analysis. When evaluated on the HAM10000 dataset, DCAN-Net achieved a precision, recall, F1-score, and accuracy of 97.00%, 97.57%, 97.10%, and 97.57%, respectively. Moreover, the application of advanced data augmentation techniques mitigated data imbalance issues and reduced false-positive and false-negative rates across the original and augmented datasets. These findings demonstrate the potential of DCAN-Net for improving clinical outcomes and advancing AI-driven skin cancer diagnostics.

A novel hybrid ViT-LSTM model with explainable AI for brain stroke detection and classification in CT images: A case study of Rajshahi region.

Computed tomography (CT) scans play a key role in the diagnosis of stroke, a leading cause of morbidity and mortality worldwide. However, interpreting these scans is often challenging, necessitating automated solutions for timely and accurate diagnosis. This research proposed a novel hybrid model that integrates a Vision Transformer (ViT) and a Long Short Term Memory (LSTM) to accurately detect and classify stroke characteristics using CT images. The ViT identifies essential features from CT images, while LSTM processes sequential information generated by the ViT, adept at capturing crucial temporal dependencies for understanding patterns and context in sequential data. Moreover, our approach addresses class imbalance issues in stroke datasets by utilizing advanced strategies to improve model robustness. To ensure clinical relevance, Explainable Artificial Intelligence (XAI) methods, including attention maps, SHAP, and LIME, were incorporated to provide reliable and interpretable predictions. The proposed model was evaluated using the primary BrSCTHD-2023 dataset, collected from Rajshahi Medical College Hospital, achieving top accuracies of 73.80%, 91.61%, 93.50%, and 94.55% with the SGD, RMSProp, Adam, and AdamW optimizers, respectively. To further validate and generalize the model, it was also tested on the Kaggle brain stroke dataset, where it achieved an impressive accuracy of 96.61%. The proposed ViT-LSTM model significantly outperformed traditional CNNs and ViT models, demonstrating superior diagnostic performance and generalizability. This study advances automated stroke diagnosis by combining deep learning innovations, domain expertise, and enhanced interpretability to support clinical decision-making, providing reliable diagnostic solutions.

C MAL: cascaded network-guided class-balanced multi-prototype auxiliary learning for source-free domain adaptive medical image segmentation.

Source-free domain adaptation (SFDA) has become crucial in medical image analysis, enabling the adaptation of source models across diverse datasets without labeled target domain images. Self-training, a popular SFDA approach, iteratively refines self-generated pseudo-labels using unlabeled target domain data to adapt a pre-trained model from the source domain. However, it often faces model instability due to incorrect pseudo-label accumulation and foreground-background class imbalance. This paper presents a pioneering SFDA framework, named cascaded network-guided class-balanced multi-prototype auxiliary learning (C MAL), to enhance model stability. Firstly, we introduce the cascaded translation-segmentation network (CTS-Net), which employs iterative learning between translation and segmentation networks to generate accurate pseudo-labels. The CTS-Net employs a translation network to synthesize target-like images from unreliable predictions of the initial target domain images. The synthesized results refine segmentation network training, ensuring semantic alignment and minimizing visual disparities. Subsequently, reliable pseudo-labels guide the class-balanced multi-prototype auxiliary learning network (CMAL-Net) for effective model adaptation. CMAL-Net incorporates a new multi-prototype auxiliary learning strategy with a memory network to complement source domain data. We propose a class-balanced calibration loss and multi-prototype-guided symmetry cross-entropy loss to tackle class imbalance issue and enhance model adaptability to the target domain. Extensive experiments on four benchmark fundus image datasets validate the superiority of C MAL over state-of-the-art methods, especially in scenarios with significant domain shifts. Our code is available at https://github.com/yxk-art/C2MAL .

Bi-Level Game Strategy for Virtual Power Plants Based on an Improved Reinforcement Learning Algorithm

To address the issue of economic dispatch imbalance in virtual power plant (VPP) systems caused by the influence of operators and distribution networks, this study introduces an optimized economic dispatch method based on bi-level game theory. Firstly, a bi-level game model is formulated, which integrates the operational and environmental expenses of VPPs with the revenues of system operators. To avoid local optima during the search process, an enhanced reinforcement learning algorithm is developed to achieve rapid convergence and obtain the optimal solution. Finally, case analyses illustrate that the proposed method effectively accomplishes multi-objective optimization for various decision-making stakeholders, including VPP and system operators, while significantly reducing curtailment costs associated with the extensive integration of distributed renewable energy. Furthermore, the proposed algorithm achieves fast iteration and yields superior dispatch outcomes under the same modeling conditions.