GEOGRAPHICALLY WEIGHTED LASSO (GWL) MODELING TO IDENTIFY FACTORS INFLUENCE STUNTING INCIDENTS IN SOUTH SULAWESI

Abstract

The Geographically Weighted Lasso (GWL) method is a technique that employs the Lasso approach within the Geographically Weighted Regression (GWR) model, which can also simultaneously select non-significant variables by shrinking the regression coefficient values to zero. Consequently, any variable assigned to a zero coefficient is considered statistically insignificant. In 2022, stunting remained a significant public health issue in South Sulawesi, ranking 10th nationwide with a prevalence of 27.2%. This underscores the urgent need for spatially sensitive analytical methods that can address regional heterogeneity and reveal key determinants at the district level. Notably, the application of GWL to analyze stunting in South Sulawesi using data from the Indonesian Nutrition Status Survey (SSGI 2022) represents a significant contribution that addresses an important research gap. This study aims to model stunting prevalence and identify its influential factors using GWL. The analysis yielded a tuning parameter λ = 0.04, achieving a model goodness of fit of R² = 0.957, demonstrating GWL’s effectiveness in mitigating multicollinearity. Four primary predictors of stunting emerged: low birth weight (LBW), access to safe drinking water, the human development index (HDI), and the average length of parental schooling.