Forecasting and Spatial Analysis of Sustainable Development Goals 1 and 4 Using TabNet

This study applies the Tabular Neural Network (TabNet) to forecast and spatially analyze Sustainable Development Goals (SDGs) 1 and 4, providing evidence-based support for policymaking aligned with the United Nations 2030 Agenda. The dataset includes 3,841 observations from 167 countries, combining World Bank socio-economic indicators as predictors with SDG index scores as targets. A country-based split allocated 80% of countries for training and 20% for testing, ensuring cross-country generalization. Future projections for 2025–2030 use a persistence-based scenario in which socio-economic covariates remain at their most recent values, and lagged SDG scores are recursively updated, producing conditional simulations rather than purely autoregressive forecasts. The TabNet model demonstrates strong predictive performance: SDG 1 achieves RMSE of 4.22 (train) and 3.99 (test) with R² of 0.89 and 0.92, while SDG 4 achieves RMSE of 3.89 and 3.76 with R² of 0.95 and 0.93. Choropleth mapping highlights regional disparities, with higher predicted scores in Europe and North America compared to Sub-Saharan Africa. These results indicate that TabNet, combined with spatial visualization, provides an interpretable, robust, and policy-relevant framework for monitoring global SDG progress.