Next-Destination Prediction with LSTM Mob Data

This project focuses on building a deep learning model to predict a user’s next travel destination based on their historical movement patterns. The system leverages GPS trajectory data, temporal features (time of day, day of week), and demographic information to learn mobility behavior.

We implemented a Long Short-Term Memory (LSTM) network, well-suited for sequential data, to capture spatiotemporal dependencies in user travel histories. The model predicts the next destination point (e.g., grid cell, location ID, or coordinates) given the sequence of past locations.

Key details:

Input Features: user’s past trajectory, timestamps, demographics.

Model: LSTM with embedding layers for spatial encoding.

Output: predicted next destination (location ID / coordinates).

Applications: personalized route recommendations, traffic flow prediction, urban planning, and ride-hailing optimization.