Pediatric Maxillofacial Trauma: Machine Learning Based Predictive Modeling to Identify Trauma Patterns.

AIM: To analyze the characteristics of pediatric facial trauma and predict factors influencing them through machine learning algorithms.

MATERIAL AND

METHODS

A prospective hospital-based study was carried out between January 2024 and January 2025. All patients up to 15 years of age reporting with maxillofacial trauma formed the sample. Collected data was subjected to logistic regression analysis and machine learning algorithms (Bayesian Network, CHAID and Neural Network) to determine factors influencing pediatric maxillofacial trauma.

RESULTS

Logistic regression and Bayesian Network demonstrated 92.59% accuracy, whereas CHAID and Neural Network showed an accuracy of 78.40% and 74.69%, respectively. Bayesian Network and Logistic Regression with similar accuracy showed age (21.77%), (p = 0.014) as the variable with maximum predictor importance, followed by paternal education (15.65%), (p = 0.003), maternal education (12.55%), (p = 0.040) and parental employment (9.43%), (p = 0.040). Further, Bayesian Network showed RTA (9.66%), sex (6.36%), maxillofacial fracture treatment (5.99%), other causes of trauma (5.67%), religion (2.89%) and delay in TDI treatment (2.78%) as other predictor variables. Parental employment had the maximum predictor importance (17.36%) in the Neural Network, whereas maternal education demonstrated the highest predictor importance (33.03%) in the CHAID model.

CONCLUSIONS

Age showed highest importance in predicting maxillofacial trauma when machine learning algorithms were used, followed by parental education and employment as other major predictors which suggested that children from lower socioeconomic status were more prone to trauma, as employment serves as a proxy for education.