Pure-inTention x G.I.: Purpose-driven Trip Demand Estimation Framework Considering Multi-dimensional Geospatial Information*

This study offers a crucial contribution at the intersection of social science and emerging technology, presenting a multidisciplinary framework to address post-pandemic challenges in urban mobility. By integrating multi-dimensional geospatial data with dynamic datasets, we introduce a novel framework designed to monitor and classify various trip purposes. Experimental results show the influence of differing social distancing policies on local movement patterns at the district level, revealing essential insights into regional trip purpose variations and their socio-economic impacts. To implement this multi-dimensional geospatial approach, we utilized diverse machine learning and deep learning techniques, alongside optimization strategies, to analyze data from multiple sources-credit card transactions, geodemographic profiles, traffic counts, local business data, and surveyed origin-destination flows. Techniques like Principal Component Analysis (PCA), Uniform Manifold Approximation and Projection (UMAP), K-means clustering, Random Forest, and Generative Adversarial Networks (GANs) were applied, with Tabu Search and Hill Climbing strategies for optimization. The well-trained prediction model developed in this study demonstrated accurate performance in understanding traffic patterns and their consequential trip purposes. The experiment results revealed the extent of the pandemic's impact across districts and their sensitivities to different policies. The developed framework proves beneficial not only for individual districts and stakeholders but also for comprehensive county and national strategic planning.