Data-Driven Modeling of Prosthetic Complications: A Literature Based Simulation and Machine Learning Approach in Prosthodontics

Creative Commons License

Toprak U.

European annals of dental sciences (Online), cilt.53, sa.1, ss.18-22, 2026 (TRDizin)

Özet

Purpose: To evaluate the impact of occlusion type, bruxism, and material characteristics on prosthetic complications using a simulated dataset analyzed with machine learning models. Materials and Methods: A retrospective computational study was conducted with synthetic data modeled on clinical prevalence and biomechanical literature. A dataset of 1,000 simulated patients included demographic, clinical, and prosthesis-related features. Prosthetic complication outcomes were generated using a logistic risk model. Five machine learning algorithms; logistic regression, random forest, gradient boosting, support vector machine, and adaptive boosting, were trained to predict complications. Model performance was assessed by accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve. Results: Bruxism, implant-opposing dentition, polymethyl methacrylate restorations, and a higher number of prosthetic units were associated with increased complication risk. Gradient Boosting markedly outperformed the other models, achieving the highest discrimination (area under the curve = 0.79; accuracy = 72.4%). Logistic Regression remained competitive (area under the curve = 0.66), while Support Vector Machine demonstrated poor recall. Conclusions: Simulation combined with machine learning can identify patients at elevated risk of prosthetic complications. Gradient Boosting showed predictive performance approaching the acceptable threshold for biomedical models, whereas other methods yielded modest results. Although not yet ready for clinical application, such data-driven approaches may support personalized prosthodontic planning and preventive strategies.