GeoLife Dataset

Citation Author(s):
Solmaz
Seyed Monir
The University of Washington
Dongfang
Zhao
The University of Washington
Submitted by:
Solmaz Seyed Monir
Last updated:
Mon, 03/31/2025 - 21:02
DOI:
10.21227/thw6-7y17
License:
60 Views
Categories:
Keywords:
0
0 ratings - Please login to submit your rating.

Abstract 

The rapid growth of spatiotemporal data makes trajectory modeling critical for extracting patterns from large-scale, dynamic mobility datasets. However, many existing methods face challenges with scalability and computational inefficiency. To address these challenges, we propose VecLSTM—a vectorized Long Short-Term Memory (LSTM) framework designed to improve both predictive accuracy and processing performance. VecLSTM introduces a novel dynamic vectorization layer that converts raw GPS trajectories into structured vector embeddings, enabling efficient storage, retrieval, and preprocessing. The architecture integrates convolutional layers for spatial feature extraction with LSTM networks for temporal sequence modeling, jointly learning spatial-temporal dependencies in dynamic mobility data. Additionally, It also integrates a structured metadata storage mechanism to encode spatial coordinates, timestamps, activity labels, and user identifiers, streamlining the learning pipeline. Experiments on two large-scale, real-world datasets—GeoLife and HighD—demonstrate that VecLSTM reduces training time by $74.2\%$, achieving an RMSE of $0.468$ and a weighted F1-score of $0.86.$ These results highlight VecLSTM's effectiveness in scalable and dynamic trajectory modeling for large-scale mobility systems.

Instructions: 

Dataset Files

    Files have not been uploaded for this dataset