Applying Machine-Learning Models on Lateral Cephalogram Parameters to Classify Skeletal Classes I and III Arab Orthodontic Patients, Instead of Individualized Parameters

A bstract Aims: The sagittal skeletal correlation involving the maxilla and mandible is commonly assessed utilizing the A point, Nasion, B point (ANB) angle and Wits appraisal. Recently, ethnicity-specific diagnostic values have been proposed to improve accuracy. This study investigates whether machine learning (ML) models can classify Arab orthodontic cases identified with skeletal class I or class III determined only by standard skeletal measurement indicators, without relying on individualized equations. Materials and Methods: Lateral cephalograms from 422 Arab orthodontic patients were analyzed in this study. Five supervised ML algorithms—linear discriminant analysis, support vector machine, K-nearest neighbor, random forest, and Classification and Regression tree—were developed and evaluated utilizing a 10-fold resampling technique. This study compared full-feature models with those using limited parameter sets, such as the Wits index and Sella–Nasion–B point angle measurement. Additionally, regression models were constructed to predict ANB angle using Wits, age, and gender. Results: Full-feature models achieved up to 97% accuracy, whereas reduced models maintained high performance (up to 91%) using Wits, S-N-B, and S-N-Pg angles. A marked positive association ( r = 0.55, P < 0.01) was found shared by Wits and ANB in class III patients. The best regression model ( R 2 = 0.57) predicted ANB using the formula: ANB = 4.37 + (0.47 × Wits) + (1.07 × gender) + (0.04 × age). Conclusions: ML models can effectively classify skeletal classes I and III malocclusions in Arab subjects using basic cephalometric data, offering a reliable alternative to individualized assessment tools.