<p>Geographically Weighted Regression (GWR) is a widely used spatial-statistical method that captures local variations in relationships between variables by accounting for spatial heterogeneity. However, traditional GWR approaches based on Euclidean distance-based kernels are limited in their ability to capture complex spatial structures. This study presents a novel model, the Adaptive Delaunay Topology-based Geographically Weighted Regression (ADT-GWR), which utilizes Delaunay triangulation and topological distance to construct a context-sensitive spatial weighting scheme. The efficacy of the ADT-GWR model is assessed by evaluating its performance on four benchmark datasets, which span the economic, health, and social domains in the United States—at different spatial scales, from neighborhood to national levels. The results demonstrate significant improvements in prediction accuracy, reduction of residual bias, and mitigation of spatial autocorrelation in terms of metrics such as RMSE, AIC, MAPE, R2, Adjusted R2, and Moran’s I compared to standard OLS and conventional GWR methods. In addition, comparisons with state-of-the-art models such as Multiscale GWR (MGWR) and Similarity GWR (SGWR) indicated a promising explanatory strength. These findings underscore the promise of topology-based spatial modeling frameworks in enhancing local regression models. The method is also available as an open source Python program.</p>