Effective decision-making in geographically dispersed supply chains depends on the ability to predict future demand patterns across interconnected regions. In multiregional settings, forecasting errors can cascade across regions, increasing uncertainty in production planning, inventory allocation, and distribution coordination. Despite advances in time series and machine learning methods, many existing approaches fail to jointly capture spatial interdependence, nonlinear temporal dynamics, and exogenous influences within a unified predictive framework, limiting their value for decision support. This study develops hybrid predictive analytics models that integrate the Generalized Space Time Autoregressive model with exogenous variables (GSTARX) and machine learning techniques to enhance multiregional demand forecasting. Three hybrid models are proposed, namely GSTARX Support Vector Regression (GSTARX-SVR), GSTARX Elman Recurrent Neural Network (GSTARX-ERNN), and GSTARX Jordan Recurrent Neural Network (GSTARX-JRNN), each capturing spatial dependence, nonlinear temporal demand dynamics, and external drivers within a unified framework. The models are evaluated using monthly cement demand data from four interconnected regions in Indonesia and benchmarked against conventional time series, exogenous-augmented, and spatial statistical models. The results show that all hybrid models improve predictive performance, with GSTARX-ERNN achieving the highest accuracy with a mean absolute percentage error (MAPE) of 15.24 percent. The hybrid framework reduces forecasting uncertainty across regions by explicitly modeling interregional interactions and nonlinear demand patterns that standalone models do not capture. These improvements enhance the reliability of production scheduling, inventory positioning, and distribution planning decisions under demand uncertainty, indicating that hybrid spatial time series and machine learning models provide practical decision support for multiregional supply chain planning and coordination. • Develop hybrid predictive analytics models for multiregional demand forecasting. • Integrate spatial dependence with nonlinear temporal learning for decision support. • Improve forecasting accuracy across interconnected supply chain regions. • Reduce demand uncertainty in production, inventory, and distribution planning. • Support operational and strategic decisions in geographically dispersed supply chains.
Prastuti et al. (Wed,) studied this question.