A deep learning framework integrating SMOTE algorithm and GC e-nose for tracing the geographical origins of food: taking Astragali Radix as an example.

Preoperative frailty for predicting in-hospital mortality in patients after cardiac surgery: an interpretable Machine learning model based on a retrospective multicenter cohort study.

Predicting PROTAC degradation activity and selectivity of effective E3 ligase through harnessing a combination of AtomPair fingerprints and multiple machine learning algorithms.

Loyalty-SMOTE: Data synthesis algorithm for effective imbalanced data classification.

Machine Learning Prediction of Hospital Stay in Pediatric Typhoid Intestinal Perforation: Pilot Study.

Birth weight is a reliable indicator of intrauterine growth and an important predictor of neonatal survival, growth, and long-term development. Globally, approximately 15.5% of live births are low birth weight, and nearly 10% are macrosomic (high birth weight), with a substantial proportion of these cases occurring in sub-Saharan Africa. Birth weights outside the normal range of 2,500-4,000g are considered abnormal and are associated with increased risks of neonatal and maternal complications. Ethiopia is similarly affected by the growing burden of abnormal birth weight. A retrospective cross-sectional study design using secondary HDSS data collected from 2015 to 2022 was employed. This design was considered appropriate because the study aimed to develop machine learning models using routinely collected surveillance data to predict abnormal birth weight and identify associated factors, rather than to establish causal relationships. All singleton births were included, and those with missing birth weight data were excluded. Six machine learning algorithms identified from the literature were built and compared to identify the best-performing model for predicting abnormal birth weight. Prior observational studies and expert opinion were used to select the candidate features for all models. The synthetic minority oversampling technique (SMOTE) was used to manage the imbalance in the dataset. The dataset was split into training (80%, n = 9,242) and testing (20%, n = 2,311) subsets for model development and evaluation. Hyper-parametric tuning was performed using grid search combined with 10-fold cross-validation to optimize model performance and reduce over-fitting. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), accuracy, precision, F1-score, and Kappa. Feature importance analysis was done using Shapley Additive explanation (SHAP) values. The Descriptive analysis of 11,553 singleton births showed that 10.78% of the newborns had high birth weight (HBW) and 9.28% had low birth weight (LBW). The eXtreme Gradient Boosting (XGBoost) model performed best by achieving an AUC of 0.835, an accuracy of 0.72, a precision of 0.67, an F1-score of 0.63, a recall of 0.54, and a kappa of 0.52 for abnormal birth weight prediction. The feature importance analysis showed that the top predictors for the low birth weight (LBW) include maternal educational status, age at first delivery, and antenatal care (ANC) visit, while high birth weight (HBW) was strongly predicted by antenatal care (ANC) visit, maternal literacy status, age at first delivery, and maternal education. Machine learning models showed moderate performance in predicting abnormal birth weight using HDSS surveillance data. Maternal educational characteristics, age at first delivery, and ANC utilization were identified as important predictive features. However, the findings should be interpreted cautiously because the model identified predictive associations rather than causal relationships. Further studies incorporating additional maternal clinical and nutritional variables, as well as external validation datasets, are recommended to improve predictive performance and generalizability.

Same-day discharge (SDD) following bariatric surgery is becoming increasingly more common to reduce healthcare utilization. However, predictors of successful SDD vary across the literature. This study applied machine learning to identify predictors of SDD and evaluate the relative contributions of patient- and procedure-related factors. Patients undergoing sleeve gastrectomy and gastric bypass were identified from the Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program database between 2020 and 2023. Patient, procedure and operative characteristics were analyzed. Synthetic Minority Oversampling Technique was applied given that SDD represented the minority of the cases. Machine learning models including Random Forest, Naïve Bayes, Neural Network, Extreme Gradient Boosting (XGBoost), and categorical boosting (CatBoost) were developed to predict SDD. Model performance was evaluated using the area under the receiver operating characteristic curve and compared with multivariable logistic regression. Feature importance was assessed using SHapley Additive exPlanations (SHAP). A total of 768,744 patients underwent bariatric surgery, of whom 66,809 (8.7%) underwent same-day discharge (SDD). SHAP analysis identified operative duration as the strongest predictor of SDD, while baseline patient comorbidities demonstrated comparatively smaller contributions to model predictions. Among machine learning models, CatBoost demonstrated the highest predictive performance (AUC 0.80), followed by XGBoost (AUC 0.79), whereas multivariable logistic regression had the lowest predictive performance (AUC 0.50). We developed a machine learning model that outperformed logistic regression in predicting same-day discharge following bariatric surgery. Operative duration emerged as the most important predictor of discharge status, suggesting that intraoperative events may play a greater role in determining discharge status than preoperative patient comorbidities.

ObjectiveTo evaluate machine learning–based models for predicting all-cause mortality in pediatric ICU patients using comprehensive biochemical panels, with a focus on addressing missing data and class imbalance.Materials and methodsA retrospective analysis was performed on a publicly available PICU dataset comprising 8,629 patients aged 28 days to 18 years. Twenty-two biochemical variables measured on the first PICU day were analyzed. Missing values were addressed using Multiple Imputation. Lasso regression was applied for feature selection. Models were trained using 5-fold cross-validation with the Synthetic Minority Oversampling Technique (SMOTE). Two ML-based imbalance-handling strategies, stacking ensemble and data subdivision were evaluated. Pairwise DeLong tests were used to compare AUC performance across models.ResultsAmong the 8,629 included patients, there were 476 non-survivors (5.5 percent). Multiple imputation followed by SMOTE improved model performance across all algorithms. The single-model classifiers achieved AUC-ROC values of 0.82 (Random Forest), 0.79 (CatBoost), 0.83 (Extra Trees), and 0.79 (Logistic Regression). The stacking ensemble demonstrated the best overall performance, with an AUC-ROC of 0.88 and an AUC-PRC of 0.45. The data subdivision approaches also produced strong discriminative performance, achieving AUC-ROC value up to 0.83 for three-subdivision, and 0.82 for five-subdivision strategies. Calibration analysis showed that the stacking model achieved the lowest Brier score (0.04), indicating superior probabilistic accuracy compared with individual classifiers. Feature importance analyses across all MI-based models consistently highlighted coagulation markers (D-dimer, reference TT, PTT, INR), electrolytes (chloride, potassium, sodium), and metabolic and organ-dysfunction indicators (AST, ALT, creatinine) as key predictors of mortality.ConclusionsThis study demonstrates that ensemble stacking is a more effective strategy than data subdivision for addressing class imbalance in PICU mortality prediction.

Accurate prediction of drug-target interactions (DTIs) is a fundamental challenge in early-stage drug discovery, particularly in the absence of reliable three-dimensional structural information. In this study, we propose a fully sequence-based DTI prediction framework that eliminates dependence on structural data while achieving docking-comparable predictive performance. The proposed framework introduces a unified representation that systematically integrates physicochemical protein descriptors, protein 3-gram sequence motifs, and sequence-like drug encodings into a single feature space, enabling effective learning across heterogeneous models. A diverse set of machine learning, deep learning, and ensemble classifiers is evaluated under stratified five-fold cross-validation with class imbalance correction using Synthetic Minority Over-sampling Technique (SMOTE). Beyond individual models, the framework incorporates advanced ensemble strategies, including a stacking classifier that combines Random Forest, Support Vector Machine, and Logistic Regression, resulting in robust performance with ROC-AUC values exceeding 0.90 and a maximum AUC of 0.914. Importantly, the framework explicitly addresses model interpretability through feature importance analysis, revealing biologically meaningful protein sequence motifs associated with binding interactions. To further substantiate the reliability of the proposed approach, molecular docking experiments are conducted on a subset of predicted drug-target pairs, and the observed agreement between docking scores and predicted binding probabilities provides independent validation. Collectively, this study demonstrates that carefully engineered sequence-derived representations, coupled with optimized ensemble learning, constitute a scalable, interpretable, and computationally efficient alternative to structure-dependent DTI prediction methods.

Traffic accident prediction in rapidly urbanizing metropolitan regions remains a critical challenge due to the complex interplay of spatiotemporal dynamics, severe class imbalance, and the opacity of predictive models that limits actionable policy interpretation. Existing approaches tend to address these challenges in isolation—deploying graph neural networks without imbalance correction, or applying oversampling without incorporating spatial context—thereby falling short of the comprehensive decision-support capability demanded by intelligent transportation systems. This paper presents a novel integrated framework, designated SLT-SHAP, that systematically unifies spatiotemporal graph convolutional learning, Synthetic Minority Oversampling Technique (SMOTE) applied exclusively to the training partition, Long Short-Term Memory (LSTM) networks for sequential temporal dependency modeling, a Transformer encoder for long-range contextual attention across hourly traffic sequences, and SHapley Additive exPlanations (SHAP) for post-hoc model interpretability. The study employs a curated spatiotemporal dataset of 132,480 observations collected at hourly resolution across 48 administrative zones in Medan Metropolitan City, Indonesia, encompassing traffic, meteorological, infrastructural, and geospatial variables with an inherent accident class imbalance of 12.4%. Experimental results demonstrate that SLT-SHAP achieves an F1-score of 0.796, AUC-ROC of 0.963, AUPRC of 0.784, and Matthews Correlation Coefficient (MCC) of 0.783, surpassing all baseline and ablation variants. Ablation analysis confirms that each component—graph construction, SMOTE, LSTM, and Transformer—contributes independently to performance. SHAP analysis identifies congestion index, hour of day, and average speed as the three most influential predictors, with spatial heatmapping delineating persistent high-risk zones. The proposed framework offers a replicable and interpretable decision-support architecture for urban road safety analytics in the Indonesian and broader Southeast Asian metropolitan context.

To construct predictive models for the recurrence of common bile duct stones (CBDS) following endoscopic retrograde cholangiopancreatography (ERCP). This retrospective study analyzed data from 1,130 patients who were randomly divided into a training set (70%) and a test set (30%). Feature selection was performed using the boruta algorithm and multivariable logistic regression (LR), followed by addressing data imbalance through the Synthetic Minority Over-sampling Technique (SMOTE). Predictive models were developed utilizing random forest (RF), extreme gradient boosting (XGBoost), and LR. We optimized these models through random search and ten-fold cross-validation to identify the best parameters. After model development, we compared their area under the curve (AUC), accuracy, recall, precision, F1-score, and decision curve analysis (DCA) to select the most optimal model. Ultimately, the optimal model was interpreted using shapley additive explanations (SHAP). Eight risk factors were identified and used to construct the predictive model, including clinical course, stone diameter, presence of multiple stones, use of biliary stents, alcohol consumption, history of biliary tract operations, presence of CBD stenosis, and endoscopic papillary balloon dilation. The RF model outperformed XGBoost and LR in terms of AUC, accuracy, recall, precision, F1-score, and DCA. The SHAP summary plot, waterfall plot, and force plot provided both overall and local explanations of the RF model. This study successfully identifies high-risk individuals for recurrent CBDS post-ERCP and offers valuable insights for machine learning-assisted clinical decision-making.

Under greenhouse production conditions, variability in fertilization management, substrate properties, and microenvironmental factors can disrupt balanced nutrient uptake, often resulting in localized or transient multi-element nutrient imbalances. Hyperspectral sensing provides continuous and high-resolution spectral information for plant nutrient assessment. However, most existing studies focus on single-element deficiencies or simplified scenarios, which limits their applicability to complex nutritional environments encountered in practice. To address this limitation, we designed a series of single- and dual-element deficiency treatments in four cultivars of chrysanthemum (Chrysanthemum morifolium Ramat.), an important cut-flower crop whose ornamental quality is highly influenced by nutrient supply. Sampling was conducted at five key growth stages across three independent experiments, yielding a total of 615 data points. Each treatment included replicates and was confirmed based on characteristic deficiency symptoms. A hyperspectral-based qualitative classification framework was developed to assess nutrient imbalances under controlled greenhouse conditions. Results indicate that although some nutrient deficiencies exhibit similar visual or phenotypic symptoms, their hyperspectral responses are distinguishable, suggesting that hyperspectral data can capture subtle differences associated with distinct nutrient imbalance conditions. To mitigate class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied, and multiple classification models were evaluated using cross-validation. The Gradient Boosting Decision Tree (GBDT) classifier combined with SMOTE showed the most consistent performance across nutrient-recognition tasks, achieving cross-validation accuracies from 0.9191 ± 0.0401 to 0.8556 ± 0.0516, balanced accuracies from 0.9595 to 0.8447, F1 from 0.9591 to 0.8496 and testing accuracies from 0.9200 to 0.8269, balanced accuracies from 0.9167 to 0.8269, F1 from 0.9140 to 0.8244. Overall, this study presents a non-destructive hyperspectral framework for classifying multi-element nutrient imbalances and demonstrates its effectiveness under greenhouse conditions, supporting hyperspectral-based nutritional assessment in ornamental crops. Further validation across diverse genotypes, seasons, and environmental conditions is needed to confirm broader applicability and model generalizability.

With the development of Internet of Things (IoT) technology, fire detection systems based on multi-sensor fusion have become critical infrastructure to ensure public safety. Due to environmental noise and sensor heterogeneity, these systems often suffer from high rates of false alarms and missed detections. Although existing machine learning approaches have partially improved classification accuracy, their overall performance remains limited. Inspired by the cognitive mechanisms of the human brain, we developed an improved mind-linked continuous-coupled neural network (ML-CCNN) based on the existing continuous-coupled neural network (CCNN). We propose a parameter adaptation mechanism that modulates neural activations through a global threshold. We utilized the synthetic minority oversampling technique (SMOTE) to mitigate data imbalance and transformed sample feature vectors into matrices for training. Our model achieved an accuracy of 99.96% on our own dataset and 99.97% on the public Smoke Detection Dataset (SDD), which highlights ML-CCNN's potential for fire detection.

Predicting permeability using seismic attributes is challenging and fraught with great uncertainty. The inverse relationship, which acts as a bridge between permeability and seismic attributes, needs to be established accurately. Machine learning, especially Gaussian Process (GP), provides a new potential. However, its application to permeability prediction is influenced by the quantity and quality of input features. This study proposes an integrated approach for permeability prediction based on GP. The approach automatically generates 222 extended features from three elastic attributes for GP training. In addition, the synthetic minority oversampling technique is used to overcome the problem of training with imbalanced samples. With the help of feature importance measures based on Shapley values, the features that are important for permeability prediction can be preferentially selected. Validation on a dolomite reservoir in Western China illustrates the role of our approach in enhancing the performance of GP in permeability prediction.

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that is manifested by sensory abnormalities such as hypersensitivity to sound and touch. Autistic children often have problems with communication, social interaction, and behavioral patterns, which are also affected by media or cartoon characters, sometimes leading to unpredictable or dangerous behavior. Timely intervention and detection are important for enhancing development. In this study, we propose a data-driven machine learning (ML) framework to detect early autism in children. The proposed approach starts with overall data processing, which involves the handling of missing values, categorical data processing, and feature selection based on Information Gain and Pearson Correlation to identify the most important attributes. The Synthetic Minority Oversampling Technique (SMOTE) is used to overcome the imbalance between classes. The proposed model uses a stacked ensemble approach in which KNN, RF, SVM, NB, and DT are used as base learners, while Random Forest works as the meta-classifier. Hyper parameter optimization is performed to further optimize the performance of the model. The models are evaluated using accuracy, precision, recall, and F1-score. The experimental outcomes revealed that the proposed ensemble model is more effective than individual classifiers, achieving 99% accuracy on the Toddler Saudi dataset, 98% accuracy on Q-CHAT, and 99% accuracy on the Nao and fused datasets. Moreover, the Shapley explanation method is applied to determine the importance of features and the impact of these features on the model predictions. The results indicate that the proposed framework may assist healthcare workers in autism screening and decision-making, serving as a promising and efficient alternative means of early autism detection.

Comparative evaluation of machine learning algorithms for early detection of metritis in dairy cows.

Machine learning models for non-invasive endometriosis triage using a laparoscopically and histologically verified cohort.

Despite diagnosis accuracy has been much improved by depending more on deep learning for disease classification, it raises serious concerns about patient data privacy, security, and scalability. Conventional centralized deep learning approaches are vulnerable to data leaks and non-compliance with such privacy rules as GDPR and HIPAA since they rely on the aggregation of sensitive medical records. Our work presents a privacy-preserving federated learning architecture enabling cooperative model training among many healthcare institutions without exposing raw patient data, hence addressing these issues. Combining autoencoder-driven hierarchical feature extraction, the proposed method improves classification performance and guarantees low information loss. Moreover applied is a hybrid class-balancing mechanism integrating generative augmentation techniques with Synthetic Minority Over-Sampling Technique (SMote) to eliminate bias in unbalanced illness datasets, so raising sensitivity for minority-class scenarios. Maintaining computational efficiency in federated systems, experimental evaluation demonstrates that the proposed model achieves accuracy of 92.5% outperforming both classic CNN-based (87.2%) and LSTM-based (89.1%) models. Strong convergence even in non-IID distributed medical data is promised by adaptive federated averaging. Furthermore, the proposed method resists adversarial attacks, therefore enhancing the security in the surrounding practical areas. This work decreases the distance between high-accurate disease categorization and privacy-preserving artificial intelligence by offering the basis for scalable, distributed, secure medical intelligence systems. The findings help federated medical artificial intelligence to grow by proving its ability to change healthcare diagnostics while keeping regulatory compliance and data security.In this study, the Synthetic Minority Over-Sampling Technique (SMOTE) is incorporated to address the class imbalance present in the distributed medical datasets. SMOTE generates new synthetic samples for minority classes by interpolating between existing minority instances, thereby preventing model bias toward majority classes and improving classification robustness across federated clients. By enhancing minority-class representation prior to federated aggregation, SMOTE ensures that the global model learns more discriminative and balanced feature patterns.

This study aimed to predict functional capacity using wearable sensor-based spatiotemporal and kinematic gait parameters, and meta-heuristic-based machine learning approaches in patients with heart failure (HF). This cross-sectional study included 70 patients with HF and assessed functional capacity and gait parameters using the six-minute walk test and a wearable inertial sensor, respectively. The Synthetic Minority Over-Sampling Technique for Regression was employed to increase the number of instances. The optimisation of four machine learning models (XGBoost, LightGBM, Random Forest, CatBoost) was implemented using Simulated Annealing, Genetic Algorithms, Particle Swarm Optimisation (PSO) and Bayesian Optimisation. The model performance was evaluated via 10-fold cross-validation through the R-squared coefficient (R²), root mean square error (RMSE), mean absolute error (MAE), and mean squared error (MSE). The interpretability of the optimal model was investigated using SHapley Additive exPlanations (SHAP) to explain individual feature contributions. The hybrid PSO-CatBoost model provided better predictive accuracy (R2 = 0.9456; RMSE = 17.2891; MAE = 10.9230; MSE = 298.9114) as compared to other hybrid optimisation-model configurations. SHAP ranked right stride length and gait speed as the most important features. The hybrid PSO-CatBoost demonstrated high predictive accuracy in estimating functional capacity in patients with HF using gait parameters, but its potential as a clinical decision-support tool requires further validation.

BackgroundNeuropsychiatric symptoms (NPS) are common in Alzheimer's disease (AD) and mild cognitive impairment (MCI), yet their detection relies on subjective assessments. Speech features offer a promising objective biomarker for NPS, reflecting emotional and cognitive states. However, existing studies are limited in terms of scale and duration.ObjectiveThis study aims to characterize acoustic features associated with NPS in early cognitive decline using Automated Assessment Model-Mini-Mental State Examination framework, and to evaluate machine learning classifiers for identifying indicators of NPS.MethodsSpeech data from 647 clinically diagnosed AD or MCI patients were collected and split into training and test sets in a 6:4 ratio. The training set was used for feature selection and model development, while test set was used for performance evaluation. The Synthetic Minority Over-Sampling Technique was applied to address class imbalance. Twelve machine learning models were trained to classify NPS categories. The best-performing models were evaluated, and SHapley Additive exPlanations (SHAP) were used to analyze feature importance.ResultsThe ExtraTrees model outperformed the others in identifying patterns associated with NPS categories, with cross-validated AUCs ranging from 0.869 to 0.901. SHAP revealed spectral_entropy_std and kurtosis_energy as key features across multiple NPS categories.ConclusionsThis study demonstrates that short speech samples obtained during the MMSE can identify acoustic patterns associated with NPS in clinically diagnosed AD and MCI using machine learning. Given the single-center design and absence of external validation, model outputs should be interpreted as directional signals to raise clinical awareness rather than as definitive diagnostic determinations.