TraffixNet

# Integrating Behavioral Planning and Public Health Strategies to Mitigate Road Traffic Crashes This repository contains the code, models, and datasets related to the research paper titled **"Integrating Behavioral Planning and Public Health Strategies to Mitigate Road Traffic Crashes"**, authored by Weilin Nie. The project explores an innovative framework that synergizes behavioral planning and public health strategies to improve road traffic safety through intelligent systems. ## 🧠 Project Overview Road traffic crashes (RTCs) are a global public health crisis. Traditional traffic models often fall short in capturing the complex, interactive dynamics between drivers. This research proposes a **multi-modal, graph-based decision-making framework** that integrates behavioral science and public health principles to enhance traffic safety in dynamic, urban environments. ### Key Components:- **Behavioral Dynamics Graph Network (BDGN):** A graph-based framework that performs real-time inference of social interactions and latent intentions of traffic agents.- **Counterfactual Intention Refinement (CIR):** A module that refines predictions using hypothetical reasoning and probabilistic simulations to adaptively mitigate risk.- **Graph-Intent Behavioral Planner (GIBP):** Integrates intention inference and trajectory planning with constraint satisfaction and utility optimization. ## 🏗️ Architecture The system is composed of three major parts: 1. **Multimodal Encoder** Extracts spatial and temporal features from agent trajectories. 2. **Graphical Propagation Layer** Encodes inter-agent relations using message-passing and soft attention mechanisms. 3. **Utility-Driven Planning & CIR** Plans and refines behavior through constraint-aware, counterfactual simulations. For detailed architectural diagrams, refer to:- *Figure 1:* GIBP Pipeline (Page 8)- *Figure 2:* Utility Planning Framework (Page 10)- *Figure 3:* CIR Components (Page 10)- *Figure 4:* Stability-Driven Regularization (Page 13) ## 📊 Experimental Results Tested on four datasets:- **CelebA**- **LSUN**- **FFHQ**- **MNIST** ### Performance Highlights:| Model | Dataset | Accuracy | Recall | F1 Score | AUC ||-------|---------|----------|--------|----------|-----|| **Ours** | CelebA | 89.64% | 86.53% | 87.90% | 91.87% || **Ours** | MNIST | 95.03% | 92.80% | 93.47% | 95.60% | See Tables 1 & 2 (Pages 15–16) for full benchmarks. ## 🔬 Technologies - **PyTorch**- **Graph Neural Networks (GNNs)**- **Multi-agent Reinforcement Learning**- **Adversarial Generative Models**- **Distributed Training (DDP)** ## 🧪 Setup and Usage > Prerequisites:- Python 3.8+- PyTorch ≥ 1.10- CUDA-compatible GPU ```bashgit clone https://github.com/your-username/road-safety-behavioral-planning.gitcd road-safety-behavioral-planningpip install -r requirements.txtpython train.py --config configs/gibp_cir.yaml