Voice-based machine learning (ML) has attracted attention as a non-invasive approach to Parkinson’s disease (PD) screening, but reported performance can be sensitive to validation design, class imbalance, and repeated recordings from the same subjects. This study evaluates a hybrid workflow that combines hyperparameter tuning, repeated Monte Carlo resampling, and probability averaging of five classifiers for PD detection using a publicly available voice-feature dataset. K-nearest neighbor (KNN), support vector machine (SVM), Gradient Boosting (GBoost), XGBoost, and a neural network (NN) were considered. The original analysis used a 70/30 train–test split, GridSearchCV for selected classifiers, and 1,000 randomized resampling iterations. Two probability ensembles were examined: an unweighted average and an MSE-rank-weighted average. The revised manuscript explicitly distinguishes these exploratory estimates from subject-independent clinical validation and documents the methodological limitations of the original analysis. In the reported experiment, the weighted ensemble obtained an F1-score of 97.78
Predicting how well a drug molecule dissolves in fats versus water—known as its LogP value—is a key step in designing effective medicines. This property influences a drug’s ability to reach its target in the body, but accurately predicting it for structurally complex molecules remains challenging. Traditional methods often show limited accuracy, as they struggle to capture intricate molecular structures. Our study proposes a new approach termed the Multi-Strategy Message Passing Graph Neural Network (MS-MPGN), which combines two complementary strategies: one that maps the overall shape of a molecule and another that focuses on local chemical interactions. By testing MS-MPGN on a dataset of 15,230 small molecules linked to cancer-related proteins, we have achieved excellent predictive performance, with a predictive accuracy (R²) of 0.951 and an error rate (RMSE) of 0.29, demonstrating competitive performance compared with standard models like Random Forest and other graph-based methods. These findings suggest that MS-MPGN can better predict LogP, offering a clearer picture of how drugs behave in the body. This tool has the potential to accelerate drug discovery, especially for crafting new cancer treatments, by guiding researchers to design molecules with just the right balance of properties.
Healthcare research plays an important role in improving medical practice by identifying health challenges, examining public health concerns, and supporting evidence-based decision-making. Considering cross-national differences in culture, infrastructure, and politics, analyzing country-specific healthcare research patterns provides valuable insights for tailored policy formulation and reform strategies. Furthermore, when common concerns arise across countries, such studies can facilitate scientific cooperation. Accordingly, the purpose of this study is to provide a cross-country overview of research themes and sub-themes in healthcare across four countries: China, England, Japan, and Iran. A total of 5122 Scopus-indexed publications related to healthcare from 2010 to 2023 were analyzed using state-of-the-art topic modeling techniques: Latent Dirichlet Allocation (LDA), Non-negative Matrix Factorization (NMF), and BERTopic. Model performance was evaluated using quantitative metrics such as coherence, stability, and runtime efficiency, as well as qualitative assessment, where BERTopic demonstrated superior overall performance and was selected for subsequent analysis. Our analysis reveals distinct national priorities: China's focus on systemic reform and AI integration, England's persistent concern with health inequalities and cancer, Japan's emphasis on long-term care for its aging population, and Iran's concentration on healthcare quality and pandemic-related research. Temporal evolution analysis further identifies emerging and declining research themes, offering actionable insights for policymakers and researchers to guide future investments, foster targeted international collaboration, and address both shared global challenges and context-specific national needs.
COVID-19 is a highly contagious and potentially fatal illness in humans, in which the air sacs of the lungs may become filled with fluid, leading to consolidations that appear as pulmonary lesions on chest X-ray (CXR) images. Thus, CXR-based analysis has been widely explored as a non-invasive tool for COVID-19 detection, and numerous methods have been reported in the literature. The objective of this work is to propose a parallel diagnostic framework that integrates CXR images with segmented pulmonary optical flow color maps to improve detection performance. Specifically, a novel convolutional neural network (CNN) that incorporates optical flow-based representations is introduced to capture structural variations and deformations in lung regions to complement radiographic intensity information. The optical flow color maps are estimated using a fractional-order variational optical flow model based on a Charbonnier norm and level set segmentation. Finally, an ensemble learning-based CNN model is trained on both CXR images and corresponding optical flow color maps to exploit complementary features. Extensive experiments and comparisons with state-of-the-art (SOTA) methods validate the performance of the proposed framework.
The purpose of this research is to develop an eco-friendly inventory model that reduces environmental harm. It encourages the customer to use organic and non-toxic products. An EOQ model for a single item is established under uncertainty. Demand is considered price-dependent, reliability-sensitive, and interval-valued. The model also considers the impact of the customers’ memory on their sales patterns. To understand the customer memory and sales behavior mathematically, we have used the Caputo-fractional differential equation in this model. Green technology is also incorporated in our model during the manufacturing process to achieve a balance between social, economic, and ecological benefits. Investment in green technology is incorporated in the model to reduce carbon emissions, pollution, and resource depletion. Shortages are allowed, which are fully backlogged, and the model is considered under a finite time horizon. A numerical example is discussed to understand the applicability and validity of the model. The proposed mathematical model is optimized using MATLAB, and the graphical illustrations are obtained to support the study. The findings illustrate that demand uncertainty and customers’ memory strongly influence inventory decisions. It is observed that an increase in memory index indicates a reduction in memory, which leads to an increase in total cost by 5% . The increase in investment for green technology reduces environmental impact and total cost by 27% . The investment in green technology also increases reliability and attracts new customers. The main aim of this study is to provide a realistic and sustainable inventory framework. The results provide useful managerial insight for the decision-makers. These insights contribute to taking profitable decision strategies while encouraging long-term social welfare and environmental sustainability.
