Molar Loss Prediction Using Machine Learning Algorithms

Purpose: This research aims to construct predictive models for estimating the long-term fate of molars in patients with periodontitis condition. Materials and Methods: A stacked ensemble model is developed that demonstrates superior accuracy compared to several other machine learning algorithms, including Logistic Regression, Support Vector Machines, Decision Trees, K-Nearest Neighbors, Random Forests, Deep Neural Networks, Gradient Boosting, and Naive Bayes. Findings: The main outcome is the accurate prediction of molar extraction following active periodontal therapy. The combined model incorporating multi-layer neural networks and logistic regression demonstrates superior area under the curve (AUC = 0.776) for total molar loss. For molar loss attributed specifically to periodontal disease, the deep neural network alone yields the highest AUC (0.774). The ensemble model also achieves the highest accuracy. Unique Contribution to Theory, Practice and Policy: By utilizing dental patients history data from the USA, this study successfully develops and validates machine learning models for predicting molar tooth loss. The combined model offered the most consistent and accurate results and is available for use in clinical settings to assist with decision-making in periodontics.