Diabetes Mellitus (DM) is a chronic metabolic disorder and a major global health problem, with many cases undiagnosed. Early detection and effective management are essential to prevent complications. This paper presents an efficient hybrid technique that combine the Synthetic Minority Oversampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN) with ensemble learning termed (SMENN-Hybrid). Gradient Boosting was identified as the most effective ensemble method through rigorous multi-metric evaluation. The proposed approach was rigorously evaluated across five diverse datasets: PIMA India, Diabetes Prediction Dataset (DPD), Diabetes Dataset 2019, Raw Merged Dataset (RMD), and Cleaned Merged Dataset (CMD). A comprehensive multi-metric assessment considering F1-Score, ROC-AUC, and Accuracy demonstrated exceptional generalizability, with Gradient Boosting achieving a composite score of 99.93/100 and maintaining coefficients of variation below 2% across all metrics (mean F1=0.9860, ROC-AUC=0.9990, Accuracy=0.9860). 5-fold stratified cross-validation confirmed remarkable stability (overall CV < 1.65% for all metrics), while systematic ablation studies validated the essential synergy between SMOTE and ENN, showing average improvements of +16.78% in F1-Score and +29.47% in Recall over unbalanced baselines. Compared to traditional methods (Logistic Regression and Decision Tree), the proposed framework achieved consistent improvements of +2.99% average F1-Score over the best baseline, with individual dataset gains ranging from +3.25% to +3.99%. Despite 246× longer training time, inference remains practical at 2.47ms, making the approach suitable for real-time clinical deployment. The combination of high effectiveness (mean F1=0.9841), exceptional consistency (CV < 2%), and comprehensive validation across multiple datasets and evaluation dimensions positions this framework as a clinically deployable solution for diabetes detection without dataset-specific tuning, offering significant advantages for similar healthcare classification tasks.

Spontaneous preterm birth (sPTB) is a complex condition with unclear etiology, associated with increased neonatal risks. Early prediction of sPTB enables timely interventions to improve outcomes. Our study aimed to construct machine learning (ML) models to predict sPTB using multi-source data, including electronic health records (EHR) and environmental factors. This retrospective cohort study included 54132 singleton pregnancies from Wuhan Children's Hospital (Wuhan Maternal and Child Healthcare Hospital) between December 2012 and December 2022. We collected multi-source predictors including demographics, routine prenatal tests, air pollution exposure, meteorological factors, and greenness exposure, resulting in a total of 82 predictors. Extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and logistic regression (LR) models were used to construct predictive models of sPTB. Screening performance was assessed via the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). Shapley additive explanation (SHAP) value was computed to assess the importance of each feature contributing to the prediction. The XGBoost model yielded the best performance in the test set with an AUROC of 0.926 and an AUPRC of 0.502. Eosinophils percentage, albumin, uric acid, amniotic fluid pocket, and sulfur dioxide exposure during late pregnancy were identified as the most important predictors of sPTB. Our results demonstrate that combining EHR data, environmental factors, and ML methods enables highly accurate and moderately precise predictions of sPTB. While the model shows promising discriminatory power, its precision requires improvement before clinical application.

Cassava (Manihot esculenta Crantz) is a staple food and a key industrial crop across tropical regions, but traditional phenotyping for critical quality traits like dry matter content (DMC) and starch content (StC) is a laborious and low-throughput process. This study investigates the efficacy of a handheld near-infrared spectrometer device (NIRS) for the non-destructive, rapid prediction of these traits. The research methodology involved collecting spectral data from 2,236 cassava clones from 19 field trials in Brazil, using two sample types: fresh roots and mashed roots. Six spectral pre-processing methods and three machine learning algorithms—Partial Least Squares (PLS), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGB)—were evaluated to optimize predictive models. Model performance was assessed using the coefficient of determination in calibration (), the root mean squared error of calibration (), and the Kappa index to quantify the consistency of clone selection. Results show that mashed samples consistently yielded superior predictive performance across all models. Specific preprocessing methods, such as Savitzky-Golay filtering combined with Standard Normal Variate (SG + SNV) and first-derivative transformations, significantly enhanced model accuracy. Among the algorithms, PLS demonstrated the best overall performance, with high predictive accuracy ( >0.96) and low prediction errors (<1.3 for DMCo), especially with mashed samples. High Kappa index values, consistently approaching 1.0, confirmed a good alignment between NIRS-based selection and traditional phenotypic methods. This study validates a portable spectrometer as a reliable and efficient tool for high-throughput phenotyping in cassava breeding programs. The findings confirm that portable NIRS devices, when used with optimal sample preparation (mashed roots) and robust modeling (PLS), can effectively yield good predictions for plant selection. This approach can significantly accelerate breeding cycles by enabling rapid, early-stage selection decisions, thereby overcoming a major bottleneck and contributing to a more efficient and sustainable genetic improvement of cassava.

Reusing long-term data from electronic health records is essential for training reliable and effective health artificial intelligence (AI). However, intrinsic changes in health data distributions over time-known as dataset shifts, which include concept, covariate, and prior shifts-can compromise model performance, leading to model obsolescence and inaccurate decisions. In this study, we investigate whether unsupervised, model-agnostic characterization of temporal dataset shifts using data distribution analyses through Information Geometric Temporal (IGT) projections is an early indicator of potential AI performance variations before model development. Using the real-world Medical Information Mart for Intensive Care-IV (MIMIC-IV) electronic health record database, encompassing data from over 40,000 patients from 2008 to 2019, we characterized its inherent dataset shift patterns through an unsupervised approach using IGT projections and data temporal heatmaps. We trained and evaluated annually a set of random forests and gradient boosting models to predict in-hospital mortality. To assess the impact of shifts on model performance, we checked the association between the temporal clusters found in both IGT projections and the intertime embedding of model performances using the Fisher exact test. Our results demonstrate a significant relationship between the unsupervised temporal shift patterns, specifically covariate and concept shifts, identified using the IGT projection method and the performance of the random forest and gradient boosting models (P<.05). We identified 2 primary temporal clusters that correspond to the periods before and after ICD-10 (International Statistical Classification of Diseases, Tenth Revision) implementation. The transition from ICD-9 (International Classification of Diseases, Ninth Revision) to ICD-10 was a major source of dataset shift, associated with a performance degradation. Unsupervised, model-agnostic characterization of temporal shifts via IGT projections can serve as a proactive monitoring tool to anticipate performance shifts in clinical AI models. By incorporating early shift detection into the development pipeline, we can enhance decision-making during the training and maintenance of these models. This approach paves the way for more robust, trustworthy, and self-adapting AI systems in health care.

Determining the triage level of patients upon their arrival at the hospital emergency department is highly important for identifying high-risk patients and allocating resources to them. This issue can be of even greater importance in patients presenting with cardiac-related symptoms. This study was conducted to predict the triage level of patients presenting with cardiac-related symptoms using machine learning methods and to compare the performance of different approaches. This prospective study was conducted in 2024 in three main steps. In the first step, a literature review was performed, and the factors influencing patient triage levels were extracted from previous studies. Then the identified factors from the literature review were presented to experts for their opinion, and the final influential factors were determined based on their feedback. In the second step, patient data were collected from the triage unit of a specialized cardiac hospital. In the third and final step, the collected data were preprocessed and then analyzed using Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Gradient Boosting (GB) machine learning methods. After reviewing the literature and surveying experts, the confirmed factors were finally identified, and data related to 1862 patients were collected. 52% of the participants in this study were male. RF achieved the highest performance with a test accuracy of 93.57%, Cohen's Kappa of 0.82, and a weighted F1-score of 0.93, followed by GB (accuracy = 90.08%, Kappa = 0.73) and SVM (accuracy = 86.6%, Kappa = 0.66). The most influential factors on patients' triage levels included: the type of high-risk condition that elevates a patient to Level 2, need for life-saving intervention, having high-risk conditions (what condition), chief complaint, level of consciousness, and diseases. In this study, five machine learning models were utilized for the triage of patients presenting with cardiac-related symptoms. The results of the study indicated that these algorithms had a good ability to discriminate between patients with different triage levels. The Random Forest method performed slightly better than the other techniques. These techniques can be used to differentiate between low-risk and high-risk patients and to allocate resources to high-risk patients.

Travel planning is a process that allows users to obtain maximum benefit from their time, cost and energy. When planning a route from one place to another, it is an important option to present alternative travel areas on the route. This study proposes a travel route planning (TRP) architecture using a Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) model to improve both travel efficiency and environmental sustainability in route selection. This model incorporates carbon emissions directly into the route planning process by unifying user preferences, location recommendations, route optimization, and multimodal vehicle selection within a comprehensive framework. By merging environmental sustainability with user-focused travel planning, it generates personalized, practical, and low-carbon travel routes. The carbon emissions observed with TRP’s artificial intelligence (AI) recommendation route are presented comparatively with those of the user-determined route. XGBoost, Random Forest (RF), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), (Extra Trees Regressor) ETR, and Multi-Layer Perception (MLP) models are applied to the TRP model. LSTM is compared with Recurrent Neural Networks (RNNs) and Gated Recurrent Unit (GRU) models. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Normalized Root Mean Square Error (NRMSE) error measurements of these models are carried out, and the best result is obtained using XGBoost and LSTM. TRP enhances environmental responsibility awareness within travel planning by integrating sustainability-oriented parameters into the decision-making process. Unlike conventional reservation systems, this model encourages individuals and organizations to prioritize eco-friendly options by considering not only financial factors but also environmental and socio-cultural impacts. By promoting responsible travel behaviors and supporting the adoption of sustainable tourism practices, the proposed approach contributes significantly to the broader dissemination of environmentally conscious travel choices.

Abstract – The project “UPI Fraud Detection Using Machine Learning” aims to provide an intelligent, real-time security layer for UPI-based digital payments by automatically identifying suspicious transactions before they are completed. An ensemble of machine learning models (including Random Forest, XG Boost, Light GBM and Gradient Boosting) is trained on a balanced fraud-non-fraud dataset, with standardized features and careful handling of class imbalance to improve recall on rare fraudulent cases while maintaining high precision. The system is deployed as a Flask-based UPI transaction portal, where users can register, log in, initiate payments, and receive instant feedback on the fraud risk for each transaction. For every payment request, the model outputs a fraud probability; high-risk transactions are automatically blocked, logged into the user’s history, and accompanied by email alerts and downloadable CSV/PDF reports for audit and analysis. Key Words: Blockchain Technology, Cybersecurity, Web Development, Machine Learning.

Using gradient boosting regression trees to identify cost drivers of laparoscopic surgery in elderly patients

The increasing occurrences of global flood events, amidst climate change, highlight the need for hydrological data availability over large geographical domains for robust decision-making. Hydrological rating curves translate fluvial stage to streamflow and play a pivotal role in various applications, including flood inundation modeling and river geomorphology. Power law regression is found to be an appropriate proxy between stage and discharge in non-linear systems. We develop a two-tier data-driven approach to predict the rating curve parameters (α, β) across the stream networks of the CONtiguous United States (CONUS). The development of rating curve models is motivated by our interest in exploring a unifying solution for representing hydrological rating curves within large stream networks. We used HYDRoacoustics in support of the Surface Water Oceanographic Topography (HYDRoSWOT), National Hydrography (NHDPlus v2.1), and STREAM-CATCHMENT (STREAMCAT) datasets. Four empirical models, Multivariate regression, eXtreme Gradient boosting (XGBoost), Random Forest, and Support Vector regression, are compared. Tier-1 XGB models offer high accuracy in prediction (R2 = 0.70) but are only applicable at gauge sites, while Tier-2 XGB models offer a good compromise between accuracy (R2 = 0.55) and applicability across the NHDPlus stream network in CONUS. We explored the influence of channel geometry and hydrometeorological attributes on rating curve parameters across NHDPlus.

Toxicity prediction of insecticides and pesticides via machine learning approach.

Achieving cloud resource optimization with trust-based access control: A novel ML strategy for enhanced performance.

Machine learning-Powered estimation of simultaneous removal of sulfamethoxazole, 17-β Estradiol, and carbamazepine via photocatalytic degradation with M-Al@ZnO.

Evaluating the accuracy of machine learning in predicting postoperative flap complications: A meta-analysis.

Radiation pneumonitis (RP) is a major dose-limiting toxicity in concurrent chemoradiotherapy (CRT) for stage III non-small cell lung cancer (NSCLC). Existing models often analyze a single lung region and rely on a single algorithm, limiting accuracy and external validity. To develop and externally validate an attention-weighted ensemble model that integrates multi-region radiomics for individualized prediction of grade ≥2 RP after three-dimensional conformal radiotherapy (3D-CRT) or volumetric-modulated arc therapy (VMAT). We retrospectively analyzed 137 patients with stage III NSCLC from two Japanese centers (training, n=107 and external validation, n=30). 40 anatomical and dose-stratified regions (covering the gross tumor volume [GTV], peritumoral shells, normal lung sub volumes, and dose sub volumes receiving 5-60Gy) were delineated on the planning CT and dose maps. From each region, 837 radiomic features were extracted from original and wavelet-filtered images. Region-wise feature reduction (variance inflation filtering and least absolute shrinkage and selection operator, LASSO) yielded four radiomic scores (Radscore Tumor, _Lung, Dose, Shell). Five base learners (random forest (RF), gradient boosting machine (GBM), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost)) were trained on the four Radscores. Their outputs were combined using an attention-weighted stacking meta-learner (SurvBETA: Survival Boosted Ensemble with Tuned Attention) and integrated with clinical covariates into a nomogram. Discrimination, calibration, and risk-group separation were evaluated using the concordance index (C-index), calibration plots, and log-rank tests. The SurvBETA + clinical nomogram achieved a C-index of 0.87 in the training cohort and 0.83 in the external validation cohort, outperforming a clinical-only model (0.54) and a conventional average-stacking ensemble (0.65). High-risk vs. low-risk groups defined by the Kaplan-Meier curve showed clear separation in cumulative RP incidence (external cohort log-rank p<0.01), with visually acceptable calibration. Decision-curve analysis indicated higher net benefit across clinically relevant thresholds compared with comparators. An attention-weighted ensemble of multi-region radiomics features, combined with clinical factors, provided accurate and externally validated prediction of symptomatic RP after CRT for stage III NSCLC.

To develop and validate a multimodal deep-learning model for predicting postoperative vault height and selecting implantable collamer lens (ICL) sizes using anterior segment optical coherence tomography (AS-OCT) and ultrasound biomicroscopy (UBM) images combined with clinical features. West China Hospital, Sichuan University, Chengdu, Sichuan, China. Deep-learning study. 626 AS-OCT and 1309 UBM images from 209 eyes of 105 participants with ICL V4c implantation were used. Features were extracted using a convolutional neural network (ResNet50) and combined with clinical data for model training. Machine learning algorithms including Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF) were used to develop models for postoperative vault height prediction and ICL size selection. Models were validated using metrics such as mean absolute error (MAE), root mean squared error (RMSE), R2 , accuracy, sensitivity, specificity, and precision. The LightGBM, XGBoost, and RF models showed RMSE values below 150 μm, MAE values below 120 μm, and R2 values around 0.4 in predicting postoperative vault height. The LightGBM model achieved the best performance in ICL size selection, with an accuracy of 0.904, sensitivity of 0.935, specificity of 0.907, and precision of 0.873, outperforming traditional methods and nearing the performance of senior doctors. The multimodal deep-learning model significantly improved the accuracy of predicting postoperative vault height and selecting ICL sizes for ICL V4c implantation, overcoming the limitations of single-modal data analysis. Future studies should expand sample sizes and conduct multicenter validations to enhance model generalizability and clinical applicability.

Artificial intelligence in the management of hospital malnutrition: A systematic review.

A comparative analysis of emotion recognition from EEG signals using temporal features and hyperparameter-tuned machine learning techniques.

Development and validation of an interpretable machine learning model for cerebral small vessel disease risk assessment.

5G networks enable ultra-high speed, low latency, and massive connectivity for critical applications such as IoT, autonomous vehicles, and digital healthcare. However, the complexity and high traffic volume in 5G architectures increase the risk of anomalies that threaten service quality and security. This study addresses the problem by proposing a real-time anomaly detection framework based on streaming data and ensemble learning algorithms. Network traffic is processed through a stream processing platform, while ensemble models such as Random Forest, Gradient Boosting, and Voting Classifier are applied to improve detection accuracy. Experimental results show that the proposed system achieves high accuracy and low latency in detecting anomalies, including Distributed Denial of Service (DDoS) attacks and technical failures. This research contributes a scalable and effective solution to enhance 5G network security and reliability, advancing the field of cybersecurity and network analytics.

Machine learning outperforms deep learning in adhesive capsulitis diagnosis: a clinical-radiomics model bridging PD-T2 MRI and multimodal data fusion.