Abstract
BACKGROUND: Despite a comprehensive national health insurance system in South Korea, achieving universal healthcare access remains challenging. Unmet healthcare needs reflect a confluence of individual-level barriers and broader structural constraints. This study aimed to identify and compare the most influential factors determining unmet healthcare needs using multiple machine learning techniques. METHODS: This study included 1,489,560 adults aged ≥ 20 who participated in the Community Health Survey between 2016 and 2022. The variables were classified into predisposing, enabling, and need factors according to Andersen’s behavioral model. Five machine learning algorithms—Decision Tree, Random Forest, XGBoost, LightGBM, and CatBoost— were used to predict unmet healthcare needs. The model performance was evaluated using accuracy, precision, recall, and F1-score metrics alongside confusion matrices, with special emphasis on minimizing false negatives. RESULTS: LightGBM exhibited the highest accuracy (0.9222) with superior recall (0.7802) and F1-score (0.7772), demonstrating optimal performance in identifying individuals with unmet healthcare needs. Age, household income, subjective health status, frequency of alcohol consumption, education level, employment status, and walking frequency were found to be the most influential predictors of unmet healthcare needs. CONCLUSIONS: The findings provide data-driven insights that enable policymakers to develop tailored interventions addressing the needs of vulnerable populations. This machine learning framework facilitates precision public health by reducing unmet healthcare needs, enhancing healthcare equity, and optimizing healthcare resource allocation.