Prediction of unerupted canines and premolars widths in an Emirati population: development and validation of regression and machine learning models

Abstract This study aimed to develop a more accurate model for predicting the widths of unerupted canines and premolars in Emirati children, using deep learning and machine learning techniques. Dental models of 380 Emirati individuals aged 15–30 years were collected. The mesiodistal widths of permanent teeth were measured with a standardized orthodontic digital caliper. Regression models were developed using linear regression, Support Vector Regression (SVR; machine learning), and Artificial Neural Networks (ANN; deep learning). The widths of mandibular lateral incisors, central incisors, and the summed width of mandibular incisors were used as predictors. A two-tailed paired t-test was used to assess differences between measured and predicted values. Model performance was evaluated using pass rate (defined as predictions within ± 1 mm of measured values), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ). The dataset was randomly divided into training (70%), validation (20%), and test (10%) sets. A statistically significant difference ( P < 0.001) was found between the values predicted by the Tanaka–Johnston equations and the measured values. In contrast, no significant differences ( P > 0.05) were observed between the measured values and those predicted by newly derived models. The highest average pass rate (78.5%, MAE 0.66) was achieved with linear regression using one predictor (summed width of the mandibular incisors). The Tanaka–Johnston method showed limited validity in the Emirati population. Population-specific regression equations significantly improved prediction accuracy, while machine-learning approaches enhanced model stability without outperforming well-calibrated linear regression models, supporting the use of simple, interpretable models for clinically reliable mixed-dentition space analysis.