A Spatio-Temporal Approach to Sales Forecasting: Leveraging Geographic and Seasonal Data for Strategic Insights
Abstract
This study investigates the integration of temporal and spatial modeling techniques to improve sales forecast accuracy for commodities influenced by seasonal, climatic, and geographic factors. In competitive markets, businesses dealing with cold beverages, frozen treats, air conditioners, and outdoor goods require precise forecasts to optimize sales strategies. A temporal model captures seasonality and time-based dependencies, while a spatial model accounts for geographic relationships among nearby stations. These components are fused using a linear regression framework to construct a robust spatio-temporal model. Model performance is evaluated using the Mean Squared Error (MSE), showing that the integrated model consistently achieves lower errors than its individual components. To assess robustness, the proposed framework is benchmarked against advanced machine learning algorithms—Random Forest, XGBoost, and LightGBM. Comparative results and relative efficiency rankings demonstrate that the spatio-temporal hybrid performs competitively, often attaining the lowest prediction errors across multiple locations. The findings confirm the value of combining spatial–temporal dependencies with modern learning approaches for accurate sales forecasting, offering actionable insights for data-driven decision-making, inventory control, and resource allocation in dynamic, climate-sensitive markets.