DST-Predict: Predicting Individual Mobility Patterns From Mobile Phone GPS Data.

Area of Science:

• Computational Social Science
• Geographic Information Science
• Artificial Intelligence

Background:

• Understanding human mobility patterns is crucial for various applications, including public health and urban planning.
• Anomalous spatial behaviors can be identified by analyzing mobility data, which is particularly relevant during events like the COVID-19 pandemic.
• Existing methods face challenges in capturing the complex, multi-scale nature of individual movement.

Purpose of the Study:

• To develop and evaluate a novel system, Deep Spatio-Temporal Predictor (DST-Predict), for predicting individual spatial behaviors.
• To forecast future visit frequencies based on historical mobility patterns derived from GPS trace data.
• To address the inherent systematic and random components in human spatial movements.

Main Methods:

• A multi-view sequence-to-sequence model utilizing Convolutional Long-short term memory (ConvLSTM) was proposed.
• The model processes past frequent visit patterns to predict future visit patterns in a multi-step approach.
• GPS trace data from 1,464 participants in western New York was used for system training and validation.

Main Results:

• The DST-Predict system demonstrated high accuracy in predicting individuals' frequency of visits to common urban locations.
• The model effectively captures complex spatio-temporal dynamics in human mobility.
• Validation on real-world GPS data confirmed the system's predictive capabilities.

Conclusions:

• The DST-Predict system offers a robust solution for predicting individual spatial behaviors using mobility data.
• Accurate prediction of human mobility patterns can enhance our understanding of complex movement dynamics.
• The developed model shows promise for applications requiring the analysis of individual spatial movements, such as anomaly detection.