Automated machine learning predicts liver metastases in patients with early-onset gastroenteropancreatic neuroendocrine tumors.

Liver metastasis is the most common site of metastasis in pancreatic neuroendocrine tumors (PaNETs), significantly affecting patient prognosis. This study aims to develop machine learning algorithms to predict liver metastasis in PaNETs patients, assisting clinicians in the personalized clinical decision-making for treatment. We collected data on eligible PaNETs patients from the Surveillance, Epidemiology, and End Results (SEER) database for the period from 2010 to 2021. The Boruta algorithm and the Least Absolute Shrinkage and Selection Operator (LASSO) were used for feature selection. We applied 10 different machine learning algorithms to develop models for predicting the risk of liver metastasis in PaNETs patients. The model's performance was assessed using a variety of metrics, including the area under the receiver operating characteristic curve (AUC), the area under the precision-recall curve (AUPRC), decision curve analysis (DCA), calibration curves, accuracy, sensitivity, specificity, F1 score, and Kappa score. The SHapley Additive exPlanations (SHAP) were employed to interpret models, and the best-performing model was used to develop a web-based calculator. The study included a cohort of 7,463 PaNETs patients, of whom 1,356 (18.2%) were diagnosed with liver metastasis at the time of initial diagnosis. Through the combined use of the Boruta and LASSO methods, T-stage, N-stage, tumor size, grade, surgery, lymphadenectomy, chemotherapy, and bone metastasis were identified as independent risk factors for liver metastasis in PaNETs. Compared to other machine learning algorithms, the gradient boosting machine (GBM) model exhibited superior performance, achieving an AUC of 0.937 (95% CI: 0.931-0.943), an AUPRC of 0.94, and an accuracy of 0.87. DCA and calibration curve analyses demonstrate that the GBM model provides better clinical decision-making capabilities and predictive performance. Furthermore, the SHAP framework revealed that surgery, N-stage, and T-stage are the primary decision factors influencing the machine learning model's predictions. Finally, based on the GBM algorithm, we developed an accessible web-based calculator to predict the risk of liver metastasis in PaNETs. The GBM model excels in predicting the risk of liver metastasis in PaNETs patients, outperforming other machine learning models and providing critical support for developing personalized medical strategies in clinical practice.

Several machine learning classifiers were trained to predict transplantation of a liver graft. We utilized 127 variables available in the DMG dataset. We included data from potential deceased organ donors between April 2012 and January 2019. The outcome was defined as liver recovery for transplantation in the operating room. The prediction was made based on data available 12-18 h after the time of authorization for transplantation. The data were randomly separated into training (60%), validation (20%), and test sets (20%). We compared the performance of our models to the Liver Discard Risk Index. Of 13 629 donors in the dataset, 9255 (68%) livers were recovered and transplanted, 1519 recovered but used for research or discarded, 2855 were not recovered. The optimized gradient boosting machine classifier achieved an area under the curve of the receiver operator characteristic of 0.84 on the test set, outperforming all other classifiers. This model predicts successful liver recovery for transplantation in the operating room, using data available early during donor management. It performs favorably when compared to existing models. It may provide real-time decision support during organ donor management and transplant logistics.

BackgroundAccurate survival prediction is critical for optimizing clinical management in small cell lung cancer (SCLC) patients with initial brain metastasis (BM). This study aims to develop a machine learning (ML)-based prognostic model to improve survival prediction, thereby enhancing clinical decision-making and treatment outcomes.MethodsData on SCLC patients with initial BM were extracted from the Surveillance, Epidemiology, and End Results (SEER) database for the period from 2010 to 2020 and split into training (70%) and testing (30%) sets. We used COX proportional hazards regression model to select modeling features in the training set. Four ML methods, including least absolute shrinkage and selection operator (LASSO), random forest (RF), eXtreme Gradient Boosting (XGBoost), and gradient boosting machine (GBM), were used to establish prognostic models and validated by both the SEER database and patients from Harbin Medical University Cancer Hospital. Model performance was evaluated using area under the curve (AUC) values, accuracy and F1 scores. Furthermore, we conducted a prioritization of the importance of features by SHapley Additive exPlanations (SHAP) at various time points.ResultsA total of 4,227 patients were enrolled in the SEER database, with 2,958 cases allocated to the training set and 1,269 to the testing set. Based on the results of univariate and multivariate COX regression analyses, we identified age, sex, marital, primary tumor size, N stage, bone metastasis, liver metastasis, lung metastasis, months from diagnosis to therapy, surgery, radiotherapy, and chemotherapy as model features. The LASSO model outperformed other models, with AUCs of 0.771, 0.724, 0.753, 0.718 at 6 months, 1 year, 2 years, and 3 years in the testing set, and 0.801, 0.763, 0.838, and 0.900 in the external validation set. Additionally, feature importance analysis consistently identified liver metastasis (highest rank), surgery, radiotherapy, lung metastasis, and chemotherapy as key predictors across all time points.ConclusionsThe LASSO model demonstrated high accuracy in predicting survival for SCLC patients with initial BM, particularly in external validation. This model may provide valuable prognostic insights for personalized treatment strategies.

BackgroundPatients with gastroenteropancreatic neuroendocrine neoplasms (GEP-NENs) and liver metastases typically exhibit poor prognoses. However, accurate survival prediction models remain insufficient. This study aimed to develop machine learning-based models to predict the 1-year, 3-year, and 5-year overall survival in these patients.MethodsWe retrospectively analyzed patients diagnosed with GEP-NENs and liver metastases from the Surveillance, Epidemiology, and End Results (SEER) database. Patients were randomly divided into training and testing sets in a 7:3 ratio. Seven machine learning models were constructed: cox regression, lasso regression, random survival forest (RSF), extreme gradient boosting (XGBoost), decision tree, gradient boosting machine (GBM), and neural network. Model performance was evaluated using C-index, AUC, Calibration curve, Brier score, and decision curve analysis (DCA). The optimal model was further interpreted through variable importance analysis, partial dependence plots, and individual prediction plots.ResultsA total of 4,528 patients were included, with 3,165 in the training set and 1,363 in the testing set. Among the seven models, the RSF model demonstrated the best overall performance. In the training set, it achieved a C-index of 0.815, with 1-year, 3-year, and 5-year AUC values of 0.895, 0.907, and 0.905, respectively, and Brier scores of 0.121, 0.128, and 0.128. The calibration curve shows good predictive performance, while the DCA highlights its strong net benefit in clinical decision-making. In the testing set, it maintained robust performance (C-index: 0.785; AUC: 0.855/0.859/0.841). The five most influential variables in the RSF model were tumor grade, surgical intervention, tumor site, age, and histology.ConclusionThe RSF model provides a reliable tool for predicting overall survival in GEP-NEN patients with liver metastases.

BackgroundPrimary gastrointestinal melanoma (PGIM) is a rare and highly aggressive malignancy with a poor prognosis, and accurate survival prediction models are currently lacking. This study aims to analyze the clinical characteristics of PGIM and develop machine learning models to predict 1-, 3-, and 5-year overall survival (OS).MethodsWe retrospectively analyzed patients diagnosed with PGIM from January 2000 to December 2021 in the Surveillance, Epidemiology, and End Results (SEER) database. Patients were randomly split into training and testing sets at an 8:2 ratio. Six algorithms were employed to construct survival prediction models: Cox proportional hazards, least absolute shrinkage and selection operator (LASSO) regression, Random Forest, Extreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), and Neural Network. Model performance and clinical utility were comprehensively evaluated using the C-index, area under the curve (AUC), Brier score, and decision curve analysis (DCA). For the best-performing model, we further utilized variable importance ranking, SHapley Additive exPlanations (SHAP) plots, individual prediction probability plots, and single-case prediction plots for interpretability, and deployed it as an online tool for clinical use.ResultsA total of 1,060 patients were included, with 845 in the training set and 215 in the testing set. Tumors were predominantly located in the rectum and anus, with a median survival of 16 months (mean: 30.3 months). The 1-, 3-, and 5-year OS rates were 61%, 30%, and 20%, respectively. Among the six algorithms, the Random Forest survival model demonstrated superior performance: in the training set, the C-index was 0.732, with AUCs of 0.813, 0.808, and 0.840 for 1-, 3-, and 5-year OS, respectively, and Brier scores of 0.175, 0.164, and 0.130. DCA confirmed high clinical net benefit. In the testing set, the model remained robust. Model analysis identified clinical stage, age, and surgical treatment as the three most critical prognostic factors.ConclusionsThe Random Forest survival model excels in predicting OS for patients with PGIM, demonstrating strong generalizability and clinical applicability, making it a valuable tool for clinical decision-making.

ObjectiveThis study aimed to develop and validate a predictive model that integrates audiological and psychometric variables to individually predict responses to sound therapy in tinnitus patients.MethodsThis study included 342 patients with chronic subjective tinnitus who received standardized sound therapy. They were randomly split into training (70%) and validation (30%) sets. Using the training set, feature selection was performed via Least absolute shrinkage and selection operator (LASSO) regression, and independent predictors were identified by multivariate logistic regression. The key variables were used to build the machine learning model, and the optimal model was determined based on the area under the receiver operating characteristic curve (AUC), calibration degree, and decision curve analysis (DCA) performance. A nomogram was created for visualization, and SHAP (SHapley Additive exPlanations) values were applied to interpret the model.ResultsA total of 342 patients were randomized into a training set (n = 239, 70%) and a validation set (n = 103, 30%). Multivariate logistic regression identified tinnitus duration, Tinnitus Functional Index (TFI) score, and Generalized Anxiety Disorder-7 (GAD-7) score as independent risk factors for treatment non-response, while previous treatment history, residual inhibition duration, uncomfortable loudness level, and Tinnitus Acceptance Questionnaire (TAQ) score were independent protective factors. Machine learning model comparisons revealed that the random forest model achieved the highest predictive performance (AUC = 0.870), outperforming support vector machine (0.801), K-nearest neighbors (0.812), and gradient boosting (0.807) models. The model also showed good calibration and provided a positive net benefit across a wide range of threshold probabilities on decision curve analysis. SHAP-based interpretability analysis confirmed the direction and magnitude of each feature’s contribution, aligning with the multivariate regression results and enhancing the model’s clinical plausibility.ConclusionIn conclusion, the developed nomogram integrates audiological and psychometric variables to individually predict sound therapy outcomes in tinnitus patients. This model serves as a practical tool for optimizing patient selection and personalizing intervention strategies, which may ultimately improve clinical efficacy and resource allocation.

Acute exacerbation of chronic obstructive pulmonary disease (AECOPD) is a major cause of hospitalization and mortality in COPD patients. Current prediction methods rely primarily on clinical symptoms and physician experience, lacking objective and precise tools. This study aimed to integrate multiple inflammatory biomarkers to develop and compare machine learning models for predicting AECOPD, providing evidence for early intervention. This retrospective study included 763 COPD patients (443 AECOPD, 320 stable COPD), randomly divided into training (n = 534) and validation (n = 229) cohorts at a 7:3 ratio. Demographic characteristics, comorbidities, and inflammatory indices were collected, including neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio (MLR), eosinophil-to-lymphocyte ratio (ELR), and basophil-to-lymphocyte ratio. After variable selection using least absolute shrinkage and selection operator (LASSO) regression, traditional logistic regression (LR) and three machine learning models-random forest, gradient boosting machine (GBM), and support vector machine-were constructed. Model performance was evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis, with SHapley Additive exPlanations (SHAP) analysis for feature importance interpretation. The GBM model demonstrated superior performance with an area under the curve (AUC) of 0.900 (95%CI: 0.858-0.942), accuracy of 0.948, specificity of 0.952, and sensitivity of 0.944 in the validation cohort, significantly outperforming the traditional LR model (AUC = 0.870). SHAP analysis identified MLR (mean SHAP value = 0.5), NLR (0.35), and pulmonary heart disease (0.32) as the three most important predictive factors. AECOPD risk increased significantly with rising MLR and NLR values, while ELR showed a negative correlation with AECOPD risk. Decision curve analysis confirmed that the GBM model provided the highest net benefit within clinically relevant threshold ranges (0.2-0.8). The GBM model integrating multiple inflammatory indices effectively predicts AECOPD. Based on routine blood test indicators without requiring expensive additional tests, this model is particularly suitable for resource-limited primary healthcare settings, providing a precise tool for early identification and individualized treatment of AECOPD, potentially improving prognosis and quality of life for COPD patients.

Machine learning-based prediction of short-term recurrence of colorectal adenomatous polyps following EMR: model development and validation study.

This study aims to develop and validate a machine learning (ML) predictive model for assessing mortality in patients with malignant tumors and hyperkalemia (MTH). We extracted data on patients with MTH from the Medical Information Mart for Intensive Care-IV, version 2.2 (MIMIC-IV v2.2) database. The dataset was split into a training set (75%) and a validation set (25%). We used the Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify potential predictors, which included clinical laboratory indicators and vital signs. Pearson correlation analysis tested the correlation between predictors. In-hospital death was the prediction target. The Area Under the Curve (AUC) and accuracy of the training and validation sets of 7 ML algorithms were compared, and the optimal 1 was selected to develop the model. The calibration curve was used to evaluate the prediction accuracy of the model further. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) enhanced model interpretability. 496 patients with MTH in the Intensive Care Unit (ICU) were included. After screening, 17 clinical features were included in the construction of the ML model, and the Pearson correlation coefficient was <0.8, indicating that the correlation between the clinical features was small. eXtreme Gradient Boosting (XGBoost) outperformed other algorithms, achieving perfect scores in the training set (accuracy: 1.000, AUC: 1.000) and high scores in the validation set (accuracy: 0.734, AUC: 0.733). The calibration curves indicated good predictive calibration of the model. SHAP analysis identified the top 8 predictive factors: urine output, mean heart rate, maximum urea nitrogen, minimum oxygen saturation, minimum mean blood pressure, maximum total bilirubin, mean respiratory rate, and minimum pH. In addition, SHAP and LIME performed in-depth individual case analyses. This study demonstrates the effectiveness of ML methods in predicting mortality risk in ICU patients with MTH. It highlights the importance of predictors like urine output and mean heart rate. SHAP and LIME significantly enhanced the model's interpretability.

This study aims to develop a machine learning-based predictive model for patients with endometriosis, with the goal of precisely identifying key factors and reliable predictive markers that influence live birth outcomes following fresh embryo transfer. Through systematic evaluation of multiple algorithms, efforts will be made to identify the optimal model for elucidating high-risk factors affecting live birth, thereby providing a basis for formulating targeted interventions to enhance the live birth rate in this population undergoing in vitro fertilization/intracytoplasmic sperm injection (IVF/ICSI). This study adopted a retrospective cohort design and included 1836 patients with endometriosis who underwent fresh embryo transfer via in vitro fertilization/intracytoplasmic sperm injection (IVF/ICSI) at Fujian Provincial Maternity and Children's Hospital between 2018 and 2023. Participants were randomly allocated to either the training set or the validation set, with a 70:30 split (1285 in the training set and 551 in the validation set), making this an internal validation study. Independent variables were screened using the least absolute shrinkage and selection operator (LASSO) and recursive feature elimination (RFE) algorithms. For eight machine learning models, namely decision tree (DT), K-nearest neighbor (KNN), logistic regression (LR), light gradient boosting machine (LightGBM), naive Bayes model (NBM), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost), we determined the optimal hyperparameter configurations using the grid search strategy. All models were trained, and their performances were evaluated through receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and Brier score (BS). The results showed that the XGBoost model exhibited the best predictive performance and was thus selected as the final modeling solution. In addition, the feature importance analysis combined with the SHapley Additive exPlanations (SHAP) value dependency plots systematically revealed the relative contributions and influence mechanisms of key features on the model predictions. Lasso and RFE analyses identified eight predictive variables for model development. The AUC values for DT, KNN, LightGBM, LR, naive Bayes, RF, SVM, and XGBoost in the training set were 0.784, 0.987, 0.841, 0.800, 0.803,0.988, 0.799, and 0.920, while those in the test set were 0.765, 0.748, 0.801, 0.805, 0.810, 0.820, 0.807, and 0.852, respectively. XGBoost demonstrated the highest predictive performance among all models. SHAP analysis identified anti-Mullerian hormone (AMH), female age, antral follicle count (AFC), infertility duration, GnRH agonist protocol, revised American Fertility Society (rAFS) stage, normal fertilization number, and number of transferred embryos as key predictors for live birth following fresh embryo transfer in patients with endometriosis. This study developed a machine learning-based predictive model for live birth following fresh embryo transfer in patients with endometriosis and systematically evaluated the comparative performance of multiple algorithms. The XGBoost model demonstrated superior overall performance, facilitating timely and precise identification of high-risk factors influencing live birth outcomes. These findings can inform targeted interventions to improve pregnancy outcomes in women with endometriosis.

Background: This study identifies optimal radiomic machine-learning classifiers to differentiate glioblastomas (GBM) from solitary brain metastases (MET), the most common neoplasms in adults, preoperatively. Methods: Four hundred and twelve patients with solitary brain tumors (242 with GBM and 170 with solitary brain MET) were divided into training (n =227) and validation (n =185) sets. Extraction of radiomic features from preoperative magnetic resonance images of each patient was accomplished with PyRadiomics software. Twelve feature selection methods and seven classification methods were evaluated to construct optimal radiomic machine-learning classifiers in the training set that were subsequently evaluated in a validation set. The role of the classifiers in differential diagnosis was evaluated using the mean area under the curve (AUC) and relative standard deviation in percentile (RSD). Findings:In the training set, thirteen classifiers had favorable predictive performances (AUC≥0.95 and RSD ≤6). In the validation set, receiver operating characteristic (ROC) curve analysis revealed that support vector machines (SVM) least absolute shrinkage and selection operator (LASSO) (AUC, 0.90) classifiers had the highest prediction efficacy, followed by Adaboost (ADa) LASSO (AUC, 0.89), Multi-Layer Perceptron (MLP) LASSO (AUC, 0.87), and random forest (RF) LASSO (AUC, 0.87). Furthermore, the clinical performance of these radiomic machine-learning classifiers were superior to neuroradiologists in accuracy, sensitivity, and specificity. Interpretation: By employing radiomic machine-learning technology, optimal machine-learning classifiers were identified for differentiating GBM from solitary brain MET preoperatively, which could significantly augment therapeutic strategies. Funding Statement: We acknowledge financial support from the National Natural Science Foundation of China (No. 81601452), and Key laboratory of functional and clinical translational medicine, Fujian province university(JNYLC1808) Declaration of Interests: The author declares no competing interest. Ethics Approval Statement: This study was approved by the ethics committee of Beijing Tiantan Hospital.

BackgroundDistant metastasis (DM) in intrahepatic cholangiocarcinoma (ICC) is associated with poor prognosis and significantly high mortality. Therefore, developing an effective early prediction method for DM risk is crucial for tailoring personalized treatment plans and improving patient outcomes.MethodsThis study included data from eligible ICC patients collected from the Surveillance, Epidemiology, and End Results (SEER) database between 2004 and 2021. Feature selection was performed using three methods, including least absolute shrinkage and selection operator (LASSO) regression, the Boruta algorithm, and recursive feature elimination (RFE). Eight machine learning (ML) algorithms were used to develop predictive models. Model performance was evaluated and compared using metrics such as the area under the receiver operating characteristic curve (AUC), area under the precision-recall curve (AUPRC), decision curve analysis (DCA), and calibration curves. The SHapley Additive exPlanations (SHAP) method was applied to rank feature importance and interpret the final model.ResultThis study included 8536 ICC patients, including 2816 (33%) with DM. The intersection results of the three feature selection methods identified 10 predictive factors. Among the 8 ML models, the gradient boosting machine (GBM) model achieved the highest AUC (0.802), AUPRC (0.571), and accuracy (0.713), as well as the lowest Brier score (0.177), indicating a comparatively robust overall performance. Calibration curves and DCA indicated that the GBM model has good clinical decision-making capability and predictive performance. SHAP analysis identified the top 10 most relevant features, ranked by relative importance: surgery, N stage, tumor grade, T stage, tumor size, radiotherapy, tumor number, age at diagnosis, chemotherapy, and number of resected lymph nodes (LNs). Additionally, a web-based online calculator was developed to predict the risk of DM in ICC patients, available athttps://bijinzhe.shinyapps.io/icc_dm_shiny/.ConclusionThe GBM model demonstrated considerable potential in predicting the risk of DM in ICC patients. This could assist clinicians in formulating personalized treatment strategies, ultimately improving the overall prognosis of ICC patients.

In this study we investigate a CT radiomics approach to predict response to chemotherapy of individual liver metastases in patients with esophagogastric cancer (EGC). In eighteen patients with metastatic EGC treated with chemotherapy, all liver metastases were manually delineated in 3D on the pre-treatment and evaluation CT. From the pre-treatment CT scans 370 radiomics features were extracted per lesion. Random forest (RF) models were generated to discriminate partial responding (PR, >65% volume decrease, including 100% volume decrease), and complete remission (CR, only 100% volume decrease) lesions from other lesions. RF-models were build using a leave one out strategy where all lesions of a single patient were removed from the dataset and used as validation set for a model trained on the lesions of the remaining patients. This process was repeated for all patients, resulting in 18 trained models and one validation set for both the PR and CR datasets. Model performance was evaluated by receiver operating characteristics with corresponding area under the curve (AUC). In total 196 liver metastases were delineated on the pre-treatment CT, of which 99 (51%) lesions showed a decrease in size of more than 65% (PR). From the PR set a total of 47 (47% of RL, 24% of initial) lesions were no longer detected in CT scan 2 (CR). The RF-model for PR lesions showed an average training AUC of 0.79 (range: 0.74–0.83) and 0.65 (95% ci: 0.57–0.73) for the combined validation set. The RF-model for CR lesions had an average training AUC of 0.87 (range: 0.83–0.90) and 0.79 (95% ci 0.72–0.87) for the validation set. Our findings show that individual response of liver metastases varies greatly within and between patients. A CT radiomics approach shows potential in discriminating responding from non-responding liver metastases based on the pre-treatment CT scan, although further validation in an independent patient cohort is needed to validate these findings.

Anastomotic strictures (AS) frequently occurs in patients following esophageal cancer surgery, significantly affecting their long-term quality of life. This study aims to develop a machine learning model to predict high-risk AS, enabling early intervention and precise management. A total of 1549 patients underwent radical esophageal cancer surgery and were split into a training set (1084) and a validation set (465). Adaptive Synthetic Sampling (ADASYN) handled class imbalance, while Boruta and Least Absolute Shrinkage and Selection Operator (LASSO) with cross-validation refined key features. High-correlation features (r > 0.8) were assessed using variance inflation factors (VIFs) and clinical relevance. Machine learning models were trained and evaluated using area under curve (AUC), accuracy, sensitivity, specificity, calibration curves, and decision curve analysis (DCA). Shapley Additive exPlanations (SHAP) analysis improved model interpretability. Seven critical variables were finalized, including anastomotic leakage (AL), neoadjuvant therapy (NCRT), suture method (SM), endoscopic assistance (EA), white blood cell count (WBC), albumin (Alb), and Suture site (SS). The Gradient Boosting Machine (GBM) model achieved the highest AUC, with 0.886 in the training set and 0.872 in the validation set. Shapley Additive Explanations (SHAP) analysis indicated that AL, SM, and NCRT were the most significant variables for model predictions. The GBM machine learning model constructed in this study can effectively identify high-risk patients for AS following esophageal cancer surgery, offering strong support for earlier postoperative detection and precise clinical management.

BackgroundThis study aimed to develop an interpretable machine learning (ML) model for predicting plaque-induced gingivitis risk in schoolchildren using questionnaire data. To enhance the model’s interpretation, SHapley Additive exPlanations (SHAP) method was applied to analyze and explain the risk factors associated with plaque-gingivitis.Materials and methodsUsing multi-stage cluster random sampling, 1755 children aged 6–12 in Lanzhou were enrolled. Participants completed a 22-item questionnaire and underwent clinical dental examinations. The collected data were stratified and randomly divided into a training set (70%) and a testing set (30%), with an independent external validation cohort (n = 120) prospectively collected for generalizability assessment. Feature selection was performed using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Six ML algorithms—Light Gradient Boosting Machine (LightGBM), random forest (RF), logistic regression (LR), eXtreme Gradient Boosting (XGBoost), decision tree (DT), and K-nearest neighbors (KNN)—were employed to process the data. The efficacy of each algorithm was evaluated using area under the curve (AUC), sensitivity (recall), specificity, accuracy, precision, F1–score and decision curve analysis. Using the SHAP method, all predictors of gingivitis prevalence in children were ranked by importance.Results51.3% (901/1755) of the children were clinically diagnosed with plaque‑induced gingivitis. 11 key predictors were selected using LASSO regression to build the ML models. Among all models, the RF achieved the highest discrimination (training AUC: 0.991; testing AUC: 0.909), followed closely by LightGBM (training AUC: 0.970; testing AUC: 0.904). The RF model was selected as the optimal model and maintained generalizability (external validation AUC: 0.824). SHAP analysis identified key predictors ranked by importance, including brushing frequency, age, regular dental checkups, brushing time, gingival bleeding during brushing, and annual income.ConclusionAn interpretable RF model accurately stratified gingivitis risk using self-reported factors. This ML-driven strategy may reduce reliance on resource-intensive clinical examinations, supporting scalable pediatric gingivitis prevention in resource-limited settings.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12903-025-07245-y.