Machine learning and SHAP value interpretation for predicting comorbidity of periodontitis and metabolic syndrome

Background Periodontitis and metabolic syndrome (MetS) frequently coexist and share inflammatory and microbiota-related mechanisms. Diet is a modifiable determinant of both conditions, yet previous studies have relied on single-nutrient approaches and linear statistics. We used interpretable machine-learning methods to identify gut microbiota–relevant dietary factors associated with periodontitis-MetS comorbidity in a nationally representative U.S. sample. Methods Cross-sectional data from 8,694 adults (NHANES 2009–2014) with complete periodontal, metabolic, dietary and demographic information were analysed. Fifteen dietary predictors—including 14 continuous food-nutrient intakes and the Dietary Index for Gut Microbiota (DI-GM)—plus nine covariates were modelled. After preprocessing (one-hot encoding, z-score standardisation, SMOTE), six algorithms (gradient boosting machine [GBM], random forest, XGBoost, LightGBM, support-vector machine, logistic regression, multilayer perceptron) were trained on 70% of the sample and validated on 30%. Performance was assessed by AUC, accuracy, F1 and calibration. SHapley Additive exPlanations (SHAP) provided global and individual feature attribution. Results Comorbidity prevalence was 16.0%. Gradient boosting models yielded the best discrimination (GBM AUC = 0.927 [95% CI 0.915–0.939]; accuracy = 0.86; F1 = 0.71). Calibration curves and decision-curve analysis confirmed clinical utility across decision thresholds. SHAP identified processed meat and coffee intake as the strongest risk contributors, followed by lower DI-GM score, low dietary fibre and high fat intake. Protective features included higher consumption of fermented dairy, whole grains, soybeans and green tea. SHAP dependence plots revealed non-linear, dose-responsive relations, with risk plateauing beyond ≈ 5 servings/day of processed meat. Conclusions An interpretable GBM model accurately predicts periodontitis–MetS comorbidity from gut microbiota–oriented dietary data. Processed meat, coffee and overall poor DI-GM score emerge as modifiable dietary targets, while fibre- and probiotic-rich foods appear protective. Although causal inference is limited by the cross-sectional design, these findings support precision-nutrition strategies and highlight the value of explainable machine learning for integrated oral-systemic disease prevention. Prospective validation is warranted.