Tianjin Metro Ridership Analysis Code with Time-Embedded Geovisual XAI
by Chuan Chen·Updated 16d ago
14.7 KB1files
Available on 1 platform
Sign in to view source links and access this dataset
Description
Analysis code for a study applying a time-embedded geovisual explainable AI framework to metro ridership in Tianjin. The code reproduces a modeling pipeline using gradient boosting and SHAP values, with models trained separately for the years 2023, 2024, and 2025. The repository includes modules for data loading, model comparison, training, evaluation, and clustering, released under a CC-BY-4.0 license by Chuan Chen.
Use Cases
Examining the temporal evolution of built environment variable contributions based on the separate yearly models.
Clustering metro stations by their spatial features and SHAP values as described in the repository modules.
Training and evaluating gradient boosting models with station-level spatial cross-validation as outlined in the pipeline.
Generating SHAP value visualizations across both space and time for model interpretation.
Strengths
The modeling pipeline is explicitly designed to examine temporal evolution by training separate models for 2023, 2024, and 2025.
The repository includes four distinct modules for utilities, model comparison, training/evaluation, and clustering, suggesting a structured workflow.
Model validation uses a station-level five-fold spatial cross-validation and evaluation on a spatially independent hold-out set.
Limitations
The primary ridership data is subject to a data sharing agreement and is not included; only the analysis code is provided.
Column-level documentation is absent; field semantics must be inferred from the code after download.
The dataset is very small (14.7 KB), indicating it contains only code, not the underlying data records.
Provenance
Source
figshare
Collection Method
Analysis code for a study using mobile signaling-derived ridership data and constructed built environment variables.
Time Range
Models cover 2023, 2024, and 2025.
Freshness
Last updated 2026-05-21 08:30:50.
Geography
Metro stations in Tianjin, China, with features constructed within an 800-meter buffer.
The actual ridership data is not included; it is available from the data provider upon request under a separate agreement. The repository contains only the analysis code in a ZIP file.