MaxEnt-Based Spatial Modelling for Detecting High-Risk Zones of Stunting Cases: Revealing the Most Important Predictive Factors and Spatial Patterns

Abstract

—The increase in stunting rates is a global concern, with implications for communities worldwide, including Indonesia. Spatial modelling using Maximum Entropy (MaxEnt) has been widely applied in geospatial health. However, the application of MaxEnt to modelling the distribution of stunting remains constrained. This research aims to identify predictive factors and spatial patterns in the distribution of stunting using spatial modelling with MaxEnt in the Sukodono District of Sidoarjo Regency, Indonesia. The data employed in this research adhere to the MaxEnt standards, incorporating spatial points and risk factors. The spatial points are utilized as input data for spatial modelling, with 275 points, of which 193 (70%) are allocated for training and 82 (30%) for model validation or testing. The environmental layers used as risk factors included the physical environment, social economy, and resource access. The physical environment was derived from satellite imagery, the social economy from kriging interpolation, and the resources from Euclidean distance. The results from the four MaxEnt Modelling schemes showed that Modelling-1, employing logistic output with 3,192 iterations, yielded the optimal prediction accuracy. The training and testing Area Under the Curve (AUC) values were 0.8969 and 0.8421, respectively. Variables significantly affecting stunting, based on risk factors, include Normalized Difference Moisture Index (NDMI), road proximity, Land Surface Temperature (LST), place of birth, and market proximity. This research highlights MaxEnt’s ability to integrate factors to map stunting risk, which can be directly used to prioritize interventions and design efficient, targeted stunting prevention programs based on spatial risk profiles.