ThomasVortex/Interactive-Traffic-Causality

Traffic Causality Reviewer

A human-in-the-loop system for discovering congestion propagation links between freeway stations using 5-minute speed data.

Features

• —Interactive Review Interface: Label causal relationships between traffic stations
• —Impact-Based Selection: Intelligently suggests pairs that would most improve model accuracy
• —Real-time Model Updates: See AUC improvements after each review session
• —Geographic Visualization: Interactive maps showing station locations and relationships
• —Active Learning: Focuses on uncertain predictions to maximize learning efficiency

How It Works

1. 1.Event Detection: Converts speed data to binary slowdown events using median-week baseline
2. 2.Causal Analysis: Computes contingency tables and closed-form causal probability (p_c) for all station pairs
3. 3.Machine Learning: Random Forest classifier predicts causal relationships
4. 4.Human Review: Users label uncertain pairs through interactive interface
5. 5.Model Improvement: Instant retraining shows AUC improvements from user feedback

Usage

The app presents traffic station pairs for review with:

• —Interactive maps showing cause/effect locations
• —Statistical metrics (causal probability, uncertainty scores)
• —Distance and time lag information
• —Simple Yes/No/Unsure labeling interface

Your labels immediately improve the model and help discover real traffic congestion patterns!