Data Science and AI for Data-Driven Transport Planning

Transport Data Collection and Analysis (TDCA)

Robin Lovelace

Online version: itsleeds.github.io/tdca
Source: github.com/itsleeds/tdca

July 6, 2026

Introduction

• Context: new digital tools for transport planning
• Reproducible data science -> impact
• Reproducible data science -> effectively integrate AI into your work
• Excel-based analysis -> dynamic, reproducible, scalable workflows
• AI has potential to reshape every sector and organisation
  • Potential only realised by people using reproducible workflows
• Goal: explain what this means and how to get started in this space

What are Data Science and AI?

Data science: turns raw data into understanding and insight

AI: generative language models to generate code and other outputs for decision-making.

• Machine Learning: pattern recognition, prediction, and optimisation: from traffic forecasting to demand modelling.
• In transport, this means handling a wide range of data sources including datasets on transport infrastructure, behaviour and more
  • There’s more data than ever before
  • Example: OpenStreetMap data which is available worldwide (Boeing 2017).

AI for Transport Planning

• AI is transforming transport (e.g., traffic prediction, safety analysis), but its value is unlocked only when integrated into reproducible workflows.
• Why reproducible environments are the best way to leverage AI:

Key Advantage	How Reproducibility Enables This	Impact on AI & Planning
Auditability & Trust	Version control (Git) logs every step, dataset, and random seed.	Prevents black-box results; ensures public policy decisions are auditable.
AI-Assisted Coding	Workflows are text-based scripts (Python/R) rather than hidden GUI menus.	AI assistants can directly write, debug, and explain code in your editor.
Agentic Loops	Execution pipelines are programmatic, automated, and self-contained.	AI agents can run automated loops to optimize parameters and test models.
Robust Stability	Package managers (Conda/Pip) pin dependencies and library versions.	Combats the non-deterministic nature of AI to ensure consistent results.

• The key insight: AI works best when the environment is defined in code, so every iteration is captured, auditable, and repeatable.

Why Reproducibility Matters

• Reproducibility: Others (or future you) can run your code and get the exact same results.
• Excel vs. Code:
  • Excel: Steps are hidden behind point-and-click menus; difficult to audit, version, or scale.
  • Code: Every transformation is explicitly documented, version-controlled, and repeatable.
• AI + Reproducibility: AI assistants help write reproducible pipelines faster, but the pipeline must remain transparent and scripted to be verified.
• Transparency: Crucial for public policy and planning — AI-driven decisions in transport affect people’s lives.

The Power of (AI-Powered) Programming

• Flexibility: You are not limited by what a GUI dropdown menu allows.
• Scalability: The same script can process 10 rows or 10 million rows.
• Sharing & Collaboration: Open-source packages mean you stand on the shoulders of giants.
• Automation: Spend more time thinking and designing solutions, less time copy-pasting.
• AI as an Accelerator: Assistants generate boilerplate and explain errors, but understanding the code you run is still essential.
• The Learning Curve: Programming has a steeper start than point-and-click tools. The Python tutorials in this module are designed to guide you step-by-step.

Tools of the Trade

• Python: A first-class citizen in data science and AI. Versatile and powerful.
• R: Excellent for statistics and spatial analysis (Arribas-Bel 2019).
• Quarto: The multi-language publishing tool — mixes code, text, and output; renders to slides, PDF, HTML (Allaire et al. 2024).

• AI coding assistants: Tools like GitHub Copilot work directly in your editor, helping you write, understand, and debug code faster.
• The ecosystem: These tools combine — Python/R for analysis, Quarto for communication, AI assistants for productivity.

Reproducible data science ecosystem

Interactive Visualisations

• Made possible with code (e.g., Python’s Streamlit, Plotly, or R’s Shiny).
• Impact: Stakeholders can “play” with the data.
• Examples:
  • Propensity to Cycle Tool (PCT)
  • Interactive road safety maps.
• Increases the transparency and reach of research.

OD flow map example from Pamplona dataset

Making the Most of Transport Data

• Diverse Data Sources: Smart card taps, GPS trajectories, traffic counts, and crowd-sourced feeds.
• Where the value lies: Not in raw datasets, but in your processing pipeline:
  • Clean & Validate: Handling missing values, sensor failures, and anomalies.
  • Visualise: Exposing spatial patterns, bottlenecks, and travel trends.
  • Model: Predicting demand and optimizing traffic flow.
• How AI Enhances This: Automating anomaly detection, clustering travel patterns, and forecasting future demand.
• Foundation for the Future: TDCA builds the core skills for working with transport data. These scale directly into next modules like Transport Data Science (TDS).

Reproducible Workflow with AI

flowchart LR
  subgraph Source["Source"]
    A[Raw transport data<br/>sensors, GPS, smart cards]
    B[Scripts<br/>Python / R / Quarto]
  end
  subgraph AI_Layer["AI Integration"]
    C[AI coding assistant<br/>Copilot / Codex]
    D[Agentic loop<br/>iterate, refine, verify]
  end
  subgraph Output["Reproducible Output"]
    E[Interactive dashboards<br/>Streamlit / Shiny]
    F[Published slides<br/>Quarto revealjs]
    G[Version-controlled<br/>Git / GitHub]
  end
  A --> B
  B <--> C
  C <--> D
  B --> E
  B --> F
  B --> G

References

Allaire, J. J., Charles Teague, Carlos Scheidegger, Yihui Xie, Christophe Dervieux, and Gordon Woodhull. 2024. Quarto. Released. https://doi.org/10.5281/zenodo.5960048.

Arribas-Bel, Dani. 2019. “A Course on Geographic Data Science.” Journal of Open Source Education 2 (14): 42. https://doi.org/10.21105/jose.00042.

Boeing, Geoff. 2017. “OSMnx: New Methods for Acquiring, Constructing, Analyzing, and Visualizing Complex Street Networks.” Computers, Environment and Urban Systems 65: 126–39. https://doi.org/10/gbvjxq.