<![CDATA[This study develops a Spatio-Temporal Physics-Informed Neural Network (ST-PINN) framework to model contraceptive use dynamics in West Sumatra, Indonesia. Existing models often assume spatial homogeneity and time-invariant parameters, which limit their ability to reflect real-world regional disparities and temporal changes in contraceptive behavior. To address this limitation, this study proposes a hybrid modeling approach that integrates differential equation-based modeling with data-driven learning in a spatio-temporal framework. The proposed model estimates contraceptive adoption, success, and failure rates across 19 districts and cities from 2012 to 2024. This study contributes by (1) developing a spatio-temporal PINN framework for dynamic parameter estimation, (2) integrating spatial and temporal data into a mechanistic model, and (3) providing region-specific insights into contraceptive dynamics. The PINN model achieves high predictive accuracy, with RMSE values ranging from 0.04678 to 0.16132 and R² values exceeding 0.85 in several regions with stable data patterns. In contrast, the ST-PINN framework provides enhanced capability in capturing spatial heterogeneity and reveals distinct regional patterns, including consistently high adoption rates in the Mentawai Islands and notable post-2018 changes across multiple regions. However, performance variability across regions indicates the presence of unobserved local factors. These findings highlight the importance of incorporating spatial and temporal heterogeneity in demographic modeling and demonstrate the significance of the ST-PINN framework as a flexible and effective tool for capturing complex regional dynamics and supporting policy-relevant decision-making in reproductive health. Nevertheless, the absence of socio-economic and behavioral variables remains a limitation. Future research should integrate additional data sources to improve model robustness and interpretability.]]>
