Covariance-Driven Graph Embedding for Real-Time Traffic Prediction

A traffic prediction model leveraging temporal graph structures to model spatial and temporal dependencies within traffic networks. This framework utilizes covariance-driven feature analysis, eigenvalue decomposition, and attention mechanisms to capture dynamic patterns, enabling accurate and scalable traffic forecasting. Tested on real-world datasets like SZ-Taxi and Los-Loop, it outperforms traditional models in accuracy and robustness.

Project Report: Report

Project Slide Deck: Slides

Getting Started

Follow the steps below to run this project locally for development and testing.

Prerequisites

Ensure the following libraries and frameworks are installed:

PyTorch
NumPy
Matplotlib

Installation

Clone the repository:

git clone https://github.com/Vamsi995/Covariance-Temporal-GCN-for-Traffic-Forecasting.git
cd Covariance-Temporal-GCN-for-Traffic-Forecasting

Install Dependencies:
```
pip install -r requirements.txt
```

Usage

Train and evaluate the model:

python train.py --config los_loop --hidden_dim 32 --epochs 100

Run evaluations:

python evaluate.py --dataset sz_taxi --hidden_dim 32 --weights_path ../cvtgcn.pkl

Datasets

The model is validated on:

SZ-Taxi: 156 major roads in Shenzhen, traffic speeds sampled every 15 minutes.
Los-Loop: 207 sensors in Los Angeles highways, data sampled every 5 minutes.