Speed Camera Effectiveness Analysis — Interactive marimo Notebook

An interactive data analysis tool built with marimo to evaluate the effectiveness of speed cameras on provincial highways. This project demonstrates how marimo's reactive notebook environment enables interactive stakeholder communication — letting project managers and policymakers explore analysis results hands-on, rather than reviewing static reports.

Demo

Why marimo?

Feature	Jupyter Notebook	marimo
Execution model	Manual cell ordering	Reactive — change a parameter, all dependent cells update automatically
File format	JSON (.ipynb)	Pure Python (.py) — git-friendly, code-reviewable
Interactive UI	Requires ipywidgets setup	Built-in widgets (sliders, tables, buttons)
Reproducibility	Hidden state issues	Guaranteed — no stale cell outputs
Deployment	Needs Jupyter server	marimo run for app mode / marimo export html for static

In this project, marimo enables:

• Adjustable parameters (buffer distance, analysis period) that reactively recompute all results
• Clickable data tables that trigger detail cards and spatial visualizations
• Embedded interactive maps (folium) showing camera locations and accident distributions
• Dashboard-style summary cards rendered as styled HTML within the notebook

Quick Start

# Run the interactive notebook (uv will auto-install dependencies)
uv run marimo edit highway_rear_end_analysis.py

# Or run in app mode (read-only, ideal for presentations)
uv run marimo run highway_rear_end_analysis.py

Prerequisites: uv (recommended) or pip. No virtual environment setup needed — marimo's inline script metadata handles dependencies automatically.

Project Structure

speed-camera-analysis-marimo/
├── highway_rear_end_analysis.py   # Main marimo notebook
├── data/
│   ├── cameras.csv                # Real speed camera locations (public data)
│   ├── accidents_demo.csv         # Synthetic accident records
│   └── sea_points.csv             # Camera exclusion list (bad coordinates)
├── scripts/
│   └── generate_demo_data.py      # Script to regenerate synthetic accidents
└── README.md

Methodology

graph LR
    A[Camera Locations] --> B[Buffer Zones<br/>GeoPandas]
    C[Accident Records] --> D[Geocoded<br/>Accidents]
    B --> E[Spatial Join]
    D --> E
    E --> F[Before / After<br/>Rate Comparison]
    F --> G[Mann-Whitney U<br/>Significance Test]
    F --> H[DiD Net Effect<br/>vs Control Group]

Loading

1. Spatial Matching: Create buffer zones around each camera → spatial join with accident records
2. Temporal Segmentation: Split accidents into pre-installation and post-installation periods
3. Rate Comparison: Monthly accident rate before vs. after camera installation
4. Statistical Validation: Mann-Whitney U test (non-parametric, p < 0.05)
5. Control Group: Difference-in-Differences using province-wide accident trends

Tech Stack

• marimo — Reactive Python notebook for interactive analysis
• pandas / GeoPandas — Data manipulation and spatial analysis
• Shapely — Geometric operations (buffer zones)
• folium — Interactive map visualization
• SciPy — Statistical testing (Mann-Whitney U)
• matplotlib / seaborn — Static chart visualization
• uv — Fast Python package manager with inline dependency support

Data Note

• Camera locations (data/cameras.csv): Real public data from Pingtung County government — speed camera installation sites are publicly disclosed information.
• Accident records (data/accidents_demo.csv): Synthetic data generated by scripts/generate_demo_data.py. The data structure mirrors real government datasets but all accident values are entirely fictional.

The analysis framework and interactive features are production-ready and have been used for actual policy evaluation work.