Abstract
<jats:p>Postoperative complications in oral and maxillofacial surgery (OMFS), pose a significant challenge to patient recovery. Accurate prediction of such complications could enable earlier interventions and improve clinical outcomes. Advances in machine learning (ML) demonstrated potential in analysing complex data within healthcare, yet their application in OMFS remains limited. Here, the use of ML models, particularly random forest is explored, to predict postoperative complications for mandibular fractures, utilising demographic, clinical, and surgical data. With nested cross validation an accuracy of 69.32% was achieved. Stratifying predictions by confidence improved classification accuracy to 85.29%, albeit with reduced breadth of applicability. These results highlight challenges related to dataset variability and model generalisability. By conducting model optimisation and exploring feature contributions, this study lays the groundwork for applying data-driven approaches within the domain of OMFS. Future efforts should focus on expanding datasets and enhancing feature engineering.</jats:p>