Diagnostic-Driven Spatial Regression Model Selection for Identifying Stunting Determinants in West Java Province

Abstract

Stunting is a key indicator of chronic malnutrition with long-term consequences for human capital development. This study proposes a diagnostic-driven framework to identify the determinants of stunting in West Java Province and to select the most appropriate spatial regression model. Classical regression, global spatial diagnostics (Moran's I and Lagrange Multiplier tests), Local Indicators of Spatial Association (LISA), and global (Spatial Autoregressive Model and Spatial Error Model) and local (Geographically Weighted Regression) spatial regression models were sequentially evaluated using health and socioeconomic variables. Although global spatial autocorrelation was not significant, LISA identified local spatial clusters in Sukabumi, Cianjur, and Garut. Among the evaluated models, GWR achieved the best performance, with the lowest AIC and RMSE and the highest $\mathrm{R}^{2}$, indicating that stunting determinants vary geographically across districts. These findings demonstrate that diagnostic-driven model selection provides a reliable framework for identifying appropriate spatial regression models and supports geographically targeted stunting intervention strategies.