<![CDATA[Accurate spatial prediction of seismicity parameters is important for characterizing seismotectonic behavior in tectonically complex regions. This study developed a hybrid spatial interpolation framework integrating Support Vector Regression–Ordinary Kriging (SVR–OK) with Inverse Distance Weighting (IDW) residual correction for estimating seismic activity (a-value) and rock fragility (b-value) across Sumatra. A total of 238 seismicity data points derived from ISC and BMKG earthquake catalogs (Mw > 5.0) for 2000–2023 were analyzed. SVR–OK was used to capture global spatial variation through flexible empirical semivariogram modeling, while IDW-based residual correction captured local variation remaining unexplained by the primary model. Performance was evaluated using LOOCV, independent testing, RMSE, MAE, RMSEP, residual diagnostics, and Moran’s I. The Hybrid SVR–OK–IDW Residual model achieved the lowest test RMSEP, reducing errors from 0.961 to 0.837 for the a-value and from 0.166 to 0.149 for the b-value, while reducing remaining positive spatial autocorrelation. Spatial contour maps identified distinct seismotectonic zonation, particularly in the Nias–Simeulue and Mentawai–Pagai segments. Simulation experiments showed that its advantage was most pronounced for clustered spatial patterns at small to moderate sample sizes (n = 64 and 225), whereas Gaussian–OK and SVR–OK performed better for unclustered patterns. These findings demonstrate the advantage of combining SVR–OK for global spatial variation with IDW residual correction for localized variation, particularly under clustered and spatially heterogeneous conditions.]]>
