The Human Development Index (HDI) is an important indicator of human development influenced by various socioeconomic factors. The relationships between HDI and its influencing factors may exhibit nonlinear patterns that require flexible regression methods. This study aims to develop a Hybrid Spline–Kernel model using a Residual Hybrid Regression approach to model HDI across 38 provinces in Indonesia using secondary data for 2024. HDI was used as the response variable, while the Poverty Rate, Open Unemployment Rate (OUR), and Labor Force Participation Rate (LFPR) served as predictors. The proposed method combines truncated spline regression to capture global patterns with multivariate local linear kernel regression to model the residual component and capture local variations. The optimal knot points were selected using the Generalized Cross-Validation (GCV) criterion, while the optimal bandwidths were determined using Least Squares Cross-Validation (LSCV). Based on the in-sample evaluation, the Hybrid Spline–Kernel model achieved a mean squared error (MSE) of 2.15, a root mean squared error (RMSE) of 1.47, and a coefficient of determination () of 0.92. Compared with the truncated spline model, the hybrid model exhibited lower in-sample errors and a higher coefficient of determination, indicating a better fit to the observed HDI data. However, these findings do not establish superior predictive accuracy for unseen observations. Further research using cross-validation and comparisons with linear and kernel-only regression models is needed to assess the predictive benefits of the proposed approach.
Copyrights © 2026