Predicting household food consumption resilience using random forest in spatially different areas of Banten province Indonesia during COVID-19

Global food system vulnerabilities highlight the urgent need to understand household food consumption resilience. This study employs a Random Forest Regressor (RFR) to model non-linear relationships between socioeconomic determinants and resilience, using the Food Consumption Score (FCS) as a proxy. Analyzing 170 households in Banten Province, Indonesia, during the COVID-19 pandemic, the RFR model demonstrated high predictive accuracy in urban (R2 = 0.837) and peri-urban (R2 = 0.868) areas, identifying income, age, and education as primary predictors. Notably, peri-urban households exhibited the highest average FCS, suggesting superior resilience derived from a “hybrid advantage”, namely simultaneous access to urban markets and agricultural production. Conversely, the model’s predictive power was lower in rural areas (R2 = 0.380), where non-monetary factors like land access and self-production predominate. This research advances food security analysis by integrating machine learning and conceptualizing peri-urban zones as strategic buffers. These data-driven insights offer a framework for strengthening food networks and nutritional education to mitigate future systemic shocks.