Predicting Visitors' Digital Footprints Using Spatial Features and Synthetic Pedestrian Data

Abstract

Assessing digital footprints of visitors has often depended on a pedestrian count data representing the human density in urban environment. The data is commonly generated via sensor, cellular phone network, or CCTV technologies, which involves high costs and raises concerns regarding personal data protection. This study presents a novel approach for predicting visitor ratings of eating and drinking out venues in Bakirkoy, Istanbul, without relying on a real pedestrian counting system. In our work, a synthetic pedestrian count measure is generated by interpolating the overall review volume data extracted from a complete set of points-of-interest data via Google Maps API. This measure is aggregated to streets, and additional spatial measures are used to further describe the urban environment. Some street geometric features (i.e., sinuosity and street length) that showed limited predictive value in earlier studies were removed from the analysis. A random forest machine learning model was applied to predict visitors' high/low ratings, achieving a 76% F1-score. The results suggest that digital traces left by visitors can be effectively predicted through alternative spatial measures and synthetic data. The proposed approach provides a cost-effective and privacy-respecting method that can support business decision-making in the context of digital consumer behavior.