A Data-driven spatiotemporal framework for retail analytics

Abstract

Retail analytics plays a critical role in optimizing inventory, improving customer targeting, and minimizing operational costs. However, traditional models often fail to effectively capture the complex interplay between spatial and temporal dynamics in consumer behavior and sales trends [1], [13]. Static clustering methods and univariate time series forecasting approaches lack the adaptability and contextual awareness necessary for modern retail operations. To address these limitations, we propose a novel data-driven spatiotemporal framework, DSFRA (Data-Driven Spatiotemporal Framework for Retail Analytics), which integrates spatial information into temporal forecasting and enhances clustering sensitivity to both time and geography. The primary objectives of this research are: 1) To develop a Spatial-Aware LSTM (SA-LSTM) that embeds geospatial features within a temporal sequence model to improve sales forecasting accuracy. 2) To design an Adaptive Spatiotemporal DBSCAN (AST-DBSCAN) algorithm for capturing regional demand fluctuations through dynamic clustering. 3) To evaluate the effectiveness of DSFRA in enhancing prediction accuracy, delivery cost reduction, and clustering quality over conventional methods.