Mining spatial-temporal patterns from customer data to improve forecasting of customer flow across multiple sites

Using a Generalized Additive Model, this study develops a novel framework to forecast hourly customer flow across multiple locations. Analyzing two years of hourly data with over 1.4 million customer visits across 106 metro stations, the model demonstrates strong potential for accurate, multi-step ahead predictions of spatial-temporal customer traffic patterns. The model provides robust forecasts to support resource planning and service levels by identifying spatial and temporal patterns in flow dynamics. The economic insights into flow dynamics and productivity possibilities enable retail headquarters to efficiently allocate staff and inventory to satisfy customers at each store. The research contributes a cost-effective city-wide hourly customer flow forecasting technique for multi-outlet retailers such as convenience stores and fast food chains.