Evaluating the Predictive Performance of Newly Developed Machine Learning Models for Shade Selection in Zirconia and Lithium Disilicate Ceramics: An In Vitro Study

Objective To evaluate the ability of machine learning (ML) models in predicting the most suitable material and shade for ceramic crowns and to assess their predictive accuracy. Materials and Methods Ninety slices of six different shades each of lithium disilicate (IPS e.max CAD) and zirconia (IPS e.max ZirCAD Prime) were prepared and evaluated against four resin composite substrates (Shofu Block) and a white background. The color coordinates of each substrate, ceramic slice, and ceramic‐substrate assembly were recorded using a spectrophotometer. The models were trained ( n = 720) and tested ( n = 180) using the coefficient of determination ( R 2 ), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). Regression outputs were mapped to categorical ceramic classes using a CIEDE2000‐based centroid mapping function, and permutation feature importance was subsequently analyzed. Results Extra Trees achieved the highest predictive accuracy ( R 2 = 0.9891, Mean Δ E 00 = 0.42 ± 0.38), with 99.44% of predictions falling within the clinical acceptability threshold (Δ E 00 ≤ 1.8). KNN demonstrated the highest mapping accuracy (98.89%). Permutation feature analysis identified ‘Specimen_L’ as the most influential factor contributing to model performance. Conclusion ML models, particularly Extra Trees and KNN, demonstrated high predictive accuracy in determining ceramic material and shade outcomes. The performance of these models suggests a promising ability to address the underlying optical relationships between substrates and restorative ceramics. Clinical significance By integrating substrate data and target shade requirements, the developed ML models offer a potential supplementary digital framework to support clinicians in material selection. This approach may contribute to more predictable esthetic outcomes and assist in streamlining the shade‐matching process.