A Machine Learning Model Exploring Creep Performance of Dental Composites

Objectives. Viscoelastic creep behaviour of RBCs determines their dimensional stability and thus contributes to their clinical performance. However, due to complex material compositional variations and differing testing protocols, comparing and analyzing the most effective factor affecting RBC viscoelastic creep behaviour is challenging. Hence, the present study aimed to establish a robust machine learning (ML) model based on datasets for creep behaviour of commercial RBCs to analyze and predict creep behaviour of RBCs and identify critical compositional factors contributing to their performance.

Materials and methods. Compressive creep deformation and recovery were measured for 5 RBCs under dry or wet conditions using a standard creep apparatus - 2 h of 20 MPa compression and 2 h creep recovery after load removal. Each RBC’s compositional data was encoded as 17 binary indices [0 and 1], which, along with experimental data, served as inputs for ML model training. 40 ML models were considered and initially ranked based on model indicators.

Results. Results showed that creep was composition-dependent and was significantly increased with water storage (p<0.05). Of the top 8 ML models, the ExtraTrees model demonstrated exceptional predictive accuracy that successfully forecasted maximum strain (%), permanent set (%) and creep recovery (%) for an additional RBC (VisCalor) (R2=0.9999). Feature importance analysis identified ORMOCER and SiO2 as the most influential monomer and filler components, respectively, affecting creep recovery.

Significance. The developed ML model offers a powerful tool for predicting RBC creep performance and uncovering key compositional factors. With future dataset expansion, this model has the potential to guide the design of advanced dental composites with improved clinical durability.