Comparing Adaptive Kernels in Geographically and Temporally Weighted Regression for Stunting in East Nusa Tenggara, 2021–2023

Danang Ariyanto (Department of Mathematics, Universitas Brawijaya, Malang, 65145, Indonesia)
Henny Pramoedyo (Department of Statistics, Universitas Brawijaya, Malang, 65145, Indonesia)
Atiek Iriany (Department of Statistics, Universitas Brawijaya, Malang, 65145, Indonesia)
Isnani Darti (Department of Mathematics, Universitas Brawijaya, Malang, 65145, Indonesia)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 10, No 2 (2026): Zero: Jurnal Sains Matematika dan Terapan

Stunting remains a major public health challenge in East Nusa Tenggara, Indonesia. This study compares Gaussian, Bisquare, and Exponential adaptive kernels in Geographically and Temporally Weighted Regression (GTWR) to model spatiotemporal variation in stunting determinants. District-level panel data from 22 districts/municipalities during 2021-2023 were used, yielding 66 space-time observations. Models were evaluated using R², RMSE, spatial-temporal parameter tests, and a spatial-only GWR baseline. The Bisquare adaptive GTWR produced the strongest in-sample fit (R² = 98.58%; RMSE = 0.7647), improving over the Bisquare GWR baseline (R² = 46.98%; RMSE = 4.6703), although this high fit should be interpreted cautiously. GRDP per capita, poverty rate, and maternal education showed significant spatial-temporal variability, while the sanitation and clean water access variable was not significant (F₃ = 1.0938; p = 0.4771). GTWR may serve as an exploratory tool for identifying local stunting patterns.