Financial risk classification of health insurance premium affordability in Indonesia using random forest: evidence from cross-sectional data

Abstract

Health insurance premium affordability classification represents a financial risk segmentation problem, particularly in emerging markets such as Indonesia, where incomplete information and reliance on observable socioeconomic indicators complicate classification processes. Conventional classification approaches, especially logistic regression, rely on linear and additive assumptions that may not adequately capture heterogeneous and interaction-based relationships among predictors. This creates a gap in understanding how nonlinear models perform in affordability-based classification under data-constrained conditions and how their outputs can be interpreted within a financial risk framework. This study evaluates the performance of Random Forest relative to logistic regression and a single decision tree in classifying health insurance premium affordability categories using cross-sectional survey data from 148 respondents in Indonesia. Model performance is assessed using accuracy, precision, recall, F1-score, and area under the curve, with validation through a 70/30 train–test split and ten-fold cross-validation. The results show that Random Forest achieves the highest performance, with accuracy of 63.3%, precision of 0.62, recall of 0.61, F1-score of 0.61, and AUC of 0.70, outperforming the decision tree and logistic regression models. Feature importance analysis identifies monthly income, age, and number of dependents as the most influential predictors of affordability classification. These findings indicate that affordability segmentation is shaped by nonlinear relationships and interaction effects among socioeconomic variables. The study contributes by demonstrating that ensemble machine learning can improve affordability-based financial risk classification while maintaining interpretability in a cross-sectional, data-constrained emerging-market context, providing a structured basis for classification-oriented decision-making in insurance pricing.