Spatio-Temporal Customer Flow and Local External Mobility Analytics, A Multi-View CNN–LSTM Framework for Hyper-Local Retail Forecasting

• Asher Noah

  covenant University

  Author

Hyper-local retail forecasting remains challenged by nonlinear, spatiotemporal demand patterns that traditional statistical and single-view deep learning models fail to capture effectively. This study addresses the gap by proposing a Multi-View CNN–LSTM framework that integrates internal customer flow data with external mobility signals for store-level demand prediction. The methodology employs a dual-branch architecture: a spatial view using convolutional neural networks to extract local mobility patterns, and a temporal view using long short-term memory networks to model customer flow dynamics. The framework was validated on a multi-year retail dataset from hyper-local store locations, enriched with mobility indicators. Results demonstrate 89.4% forecasting accuracy, outperforming baseline LSTM (82.1%), CNN (79.6%), and seasonal ARIMA (71.3%) models. Feature importance analysis reveals that external mobility signals contributed 34.2% to predictive performance, confirming their utility as complementary predictors. The study concludes that multi-view architectures integrating heterogeneous spatiotemporal signals substantially improve hyper-local retail forecasting accuracy, offering practitioners a replicable framework for inventory optimization and staffing decisions.

• pdf

Copyright (c) 2026 Asher Noah (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.