Application of Machine Learning Approaches to Identify Factors Associated with Myopia Among Preschool Children

Abstract Objective Myopia has become a major public health concern worldwide, with an increasing trend toward earlier onset. However, factors associated with myopia among preschool children remain insufficiently characterized. This study aimed to identify potential factors associated with preschool myopia using machine learning approaches. Methods A school-based cross-sectional study was conducted among 1,935 children aged 3–7 years in Huaibei, China. Information on demographic characteristics, early life factors, family history, and behavioral exposures was collected through caregiver questionnaires after routine vision screening. Seven machine learning algorithms were applied to identify important factors associated with myopia. Results The prevalence of caregiver-reported myopia was 2.33%. Among the evaluated models, the Super Learner model demonstrated the best overall performance, with an AUC of 0.742. Variable importance analysis identified parental myopia, close distance television viewing, birth weight, breastfeeding duration, and paternal education as important factors associated with preschool myopia. These findings suggested that early childhood myopia was influenced by multiple domains, including familial, early-life, and behavioral factors. Conclusions Machine learning approaches provide a useful framework for integrating multidimensional information and identifying potential factors associated with preschool myopia. These findings may support targeted associated assessment and prevention strategies in preschool vision care.