Development and Comparison of AI Algorithms for a Predictive Model of Cataracts: Analysis of the Korea National Health and Nutrition Examination Survey (2015–2017)

Cataracts are a major cause of visual impairment worldwide, particularly among older adults, with an increasing prevalence due to population aging. Surgery is the primary treatment; however, preventive strategies are crucial for reducing the disease burden. This study aimed to investigate dietary and health-related factors associated with cataract occurrence and develop a predictive model using machine learning. Data were derived from the Korea National Health and Nutrition Examination Survey 2015-2017. The study included 190 women aged 60-79 years: 124 with cataracts and 66 controls. Analyzed variables included sociodemographic, behavioral, chronic disease, and dietary intake factors. After data preprocessing, 4 machine learning algorithms: support vector machine (SVM), random forest (RF), eXtreme gradient boosting, and multilayer perceptron were used. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC) and precision-recall curves. Among the tested models, the SVM achieved the best performance under stratified 10-fold cross-validation, with 71% accuracy, 86% precision, 73% recall, 79% F1-score, 65% AUROC, and 81% AUPRC. According to our findings, the odds of having cataracts can be effectively predicted using dietary and health data without relying on specialized ophthalmic equipment. The proposed model demonstrates the potential of machine learning-based tools for early identification and prevention of cataracts. Future studies with larger and more diverse samples, as well as integrating additional data sources such as genomics and lifestyle factors, are warranted to refine predictive accuracy and enhance personalized nutrition-based interventions.