Accurate road traffic estimation is essential for efficient traffic management and strategic urban planning. However, traditional infrastructure-based traffic sensing systems face significant limitations in providing high-resolution, network-wide coverage due to prohibitive installation and operational costs. Consequently, there is a growing need to explore alternative data sources to complement conventional traffic monitoring. To address this gap, this paper investigates and justifies the applicability of crowdsourced social activity data as a novel data source for road traffic estimation. In the era of smartphones, social activity information is widely available, with representative data accessible at the majority of Points of Interest (POIs). In this study, a relationship is established between the social activity represented by Google Popular Times (GPT) data at POIs and conventional traffic information, such as speed (from Floating Car Data) and number of vehicles (from video sensors). Spatiotemporal correlation and prediction capabilities are presented and proven using real-world data from two different metropolitan areas: Budapest (Hungary) and Kyoto (Japan). A probabilistic model is introduced, enabling the regression of traffic data on GPT using a dimensionality reduction approach. To capture temporal features, forward- and backward-shifted social activity data is also applied in the estimation methodology. The analysis justifies that only a fraction of the social activity information is enough for efficient traffic prediction (both for speed or number of vehicles), resulting in a few percent relative errors. • Probabilistic model using PCA estimates traffic solely from Google Popular Times. • Time-shifted POI data captures temporal features for generalized regression. • The approach successfully predicts both vehicle counts and fleet speed data. • Validated with real-world traffic data from Budapest and Kyoto.
Törő et al. (Sat,) studied this question.