Skip to main content
eScholarship
Open Access Publications from the University of California

UCLA

UCLA Electronic Theses and Dissertations bannerUCLA

Context-Aware Deep Learning Model for Predicting\ Non-Mandatory Activity Locations

Abstract

The explosion of mobile internet usage has generated vast amounts of data on users' spatiotemporal activities. This data is crucial for studying human movement, enhancing traffic management, and understanding epidemic spread dynamics. Our study aims to tackle the challenge of predicting locations for non-mandatory activities using mobile location data.

The thesis introduces the SageGRU model, which combines GraphSage with Attentional Gated Recurrent Units, to predict the next visiting Point of Interest (POI) for non-mandatory activities. SageGRU leverages historical visitation data, categorizes activity types, and considers temporal dimensions and global movement trends using POI-to-POI transition graphs. Validated with the Veraset dataset, SageGRU achieves 10.2\% accuracy at predicting the top location, 20.8\% for the top two locations, and 27.4\% for the top three, significantly outperforming existing models. This highlights the importance of comprehensive spatio-temporal context in predicting non-mandatory activity locations. SageGRU’s capability to reconstruct real-life trajectories and aggregate travel patterns underscores its potential to advance urban mobility and public health planning by offering deeper insights into human movement.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View