Context-Aware Mobility Modeling for Next-Location Prediction Using STC-LSTM

Abstract

Accurate next-location prediction remains a challenging problem in spatio-temporal mobility modeling. Many existing approaches rely primarily on raw trajectory sequences and overlook the semantic and contextual factors that govern human movement behavior. To address this limitation, this study proposes a novel Spatio-Temporal Context Long Short-Term Memory (STC-LSTM) framework that explicitly integrates contextual information into next-location prediction. The framework enriches dense trajectory data with temporal and spatial semantics, including day of the week, holiday indicators, land-use categories, and postal codes. These contextual features are transformed through embedding representations and combined with spatial-temporal trajectory sequences, enabling the model to learn semantic relationships and long-term mobility dependencies. The resulting representations are processed through stacked LSTM layers with dropout regularization and dense prediction layers to capture mobility patterns and forecast future locations. Experiments are conducted on a large-scale real-world dataset comprising approximately 56,000 trajectories and 5.6 million Global Positioning System (GPS) points, following a pipeline that includes data cleaning, map-matching, contextual annotation, normalization, and LSTM-based modeling. Results demonstrate that incorporating contextual information substantially improves prediction performance over trajectory-only baselines, yielding accuracy improvements from approximately 38% to over 75%. The best-performing configuration, integrating all contextual features, achieves a minimum mean squared error (MSE) of 1.04 × 10−5 , compared to 7.80 × 10−5 for the no-context baseline. Furthermore, spatial evaluation shows a reduction in distance error from over 1.5 km in the baseline model to approximately 384 m, demonstrating the effectiveness of STC-LSTM for accurate and robust real-world next-location prediction.