Machine Learning Models for Early Gingival Health Screening Using Routine Clinical and Behavioral Data

Abstract

This study addresses the high prevalence of periodontal disease in Indonesia by developing an accessible, early screening tool that overcomes the limitations of existing AI approaches, which heavily rely on costly radiographic imaging. We developed a machine learning framework utilizing demographic, clinical, behavioral, and medical-history data from 500 patients at the Maninjau Community Health Centre. Four algorithms—Random Forest, Logistic Regression, Support Vector Machine (SVM), and XGBoost—were evaluated using stratified 5-fold cross-validation and ROC curve assessment. The results demonstrate that Random Forest achieved the highest overall classification performance, attaining an accuracy of 95.2%, an AUC of 0.983, and an F1-score of 0.9521, with superior capability across all categories (Normal, Gingivitis, and Periodontitis). To demonstrate practical applicability, this model was deployed as a Streamlit-based prototype decision-support application for real-time clinical use. This research provides significant value by demonstrating that routinely collected, non-invasive clinical and behavioral data can serve as potent predictors for periodontal screening, eliminating the dependency on expensive imaging. This offers a cost-effective digital health framework suitable for primary healthcare settings. Future research will focus on validating this model using larger, multi-center datasets to ensure broader generalizability.