Integrating Conventional and Machine Learning Approaches for Analyzing Pain Experience in Adolescents with Temporomandibular Disorders: A Comparative Study

Purpose: Temporomandibular disorders (TMD) are multifactorial conditions influenced by pain severity (PS), psychosocial distress, and sleep disturbances.This study integrates conventional regression models and machine learning (ML)-based decision trees to enhance understanding of pain interference in adolescents with TMD.Methods: A retrospective analysis of 662 patients were conducted using clinical data and self-reported questionnaires: the Brief Pain Inventory, the Pain Catastrophizing Scale, the Symptom Cheklist-90-Revised, the Pittsburgh Sleep Quality Index.Linear and logistic regression models identified key predictors, while decision tree models stratified patient subgroups based on threshold-based classifications.Model performance was assessed.Results: Both regression and decision tree models identified PS and pain catastrophizing as the model critical predictors of pain interference.Regression models quantified statistical significance and effect sizes, explaining 65% of variance.Decision trees provided hierarchical stratifications, revealing that patients with high PS and catastrophizing experienced significantly greater pain interference.Sleep quality influenced subgroups but was not a primary splitting factor.Decision trees achieved an R 2 of 0.54 for regression and 0.26 for classification, indicating moderate predictive power.Conclusions: Conventional regression models provide inferential clarity, while ML-based decision trees enhance clinical interpretability through subgroup identification.Combining these approaches optimizes patient stratification and supports personalized treatment strategies.