<![CDATA[The Rapid Assessment of Avoidable Blindness (RAAB) surveys provide crucial information for planning and evaluating eye health initiatives, particularly in low- and middle-income countries where data systems are often limited. RAAB results are analyzed to estimate the prevalence of visual impairment and to assess cataract surgical coverage across populations. However, despite their rich individual-level data, RAAB surveys have rarely been explored for predictive modeling that could proactively identify people most vulnerable to blindness. This study sought to address that gap by developing and validating interpretable machine-learning models capable of predicting individuals at the highest risk of avoidable blindness. We used RAAB 2016 data collected from seven provinces across Indonesia, comprising a large and diverse sample of older adults. Two modeling approaches—a calibrated Extreme Gradient Boosting (XGBoost) algorithm and a Logistic LASSO regression—were trained and evaluated. Both models demonstrated outstanding discrimination (AUC ≈ 0.96) and strong calibration performance (Brier score ≈ 0.02), ensuring that predictions corresponded well to actual outcomes. Key predictors consistently selected across methods included increasing age, presence or absence of lens opacity, self-reported functional difficulty in seeing or mobility, and lack of corrective spectacles. To enhance usability in field settings, we also derived a simplified point-score tool from the LASSO model. Decision-curve analysis confirmed that the model could offer substantial clinical and operational benefit by guiding targeted outreach where resources are limited. Overall, this work highlights predictive analytics as promising extension of the RAAB framework, enabling more precise and efficient public eye health strategies in Indonesia.]]>
