Systematic comparison of machine learning algorithms to develop and validate predictive models for periodontitis.

Area of Science:

• Computational biology
• Dental informatics
• Machine learning in healthcare

Background:

• Periodontitis is a prevalent inflammatory disease affecting tooth-supporting structures.
• Accurate prediction of periodontitis is crucial for timely intervention and management.
• Machine learning (ML) offers potential for developing sophisticated predictive models.

Purpose of the Study:

• To evaluate and compare the predictive performance of various machine learning algorithms for periodontitis.
• To assess the validity of ML models using both internal and external validation strategies.
• To investigate the impact of data pre-processing techniques on model performance.

Main Methods:

• Utilized national survey data from Taiwan and the United States (total n=7138).
• Trained 10 distinct machine learning algorithms to predict periodontitis.
• Validated models internally (bootstrap) and externally (cross-country datasets), assessing six performance metrics.

Main Results:

• Algorithms demonstrated excellent performance during internal validation (AUC > 0.95, accuracy > 95%).
• External validation revealed a significant drop in predictive performance across all models.
• Model performance varied based on data pre-processing methods and the training cohort.

Conclusions:

• Current machine learning models show limited generalizability for periodontitis prediction.
• Enhanced predictive accuracy requires larger sample sizes and more intricate predictor variables.
• Further research is needed to optimize ML algorithms for robust periodontitis risk assessment.