🚑 AI-Powered Arabic Voice Assistant for Real-Time First Aid

📌 Project Overview

In medical emergencies, every second counts. This project introduces an AI-powered Arabic Voice Assistant designed to provide immediate, step-by-step first aid guidance. By leveraging Speech Recognition and Natural Language Processing (NLP) through the AraBERT model, the system interprets user descriptions of emergencies and delivers medically verified instructions in real-time.

🚀 Key Features

• Voice-to-Text Integration: Users can describe the injury in natural Arabic speech.
• Intelligent Classification: Uses a fine-tuned AraBERT model to categorize injuries and assess their severity (Low, Medium, High).
• Medically Verified Content: Instructions are based on official guidelines from the Saudi Ministry of Health.
• Arabic Dialect Support: Fine-tuned to understand various ways people describe emergencies in Arabic.

🧠 Technical Methodology

The project follows a rigorous Machine Learning pipeline:

1. Data Engineering: Preprocessing a custom dataset of 700+ records, including scenarios for burns, wounds, and nosebleeds.
2. Composite Labeling: Training the model on a combination of Injury Type + Severity for more granular advice.
3. Model Fine-tuning: Utilizing the aubmindlab/bert-base-arabertv2 transformer model for deep linguistic understanding.
4. Optimization: Implementing stratified splits and advanced evaluation metrics (Precision, Recall, F1-Score).

📊 Performance Results

The model achieved an impressive 88% Accuracy overall.

• High Performance: Excellent recall for critical injuries like severe burns and wounds.
• F1-Score: Strong balance between precision and recall across most emergency categories.

🛠️ Tech Stack

• Language: Python
• Libraries: Transformers (Hugging Face), Scikit-learn, Pandas, NumPy, PyTorch.
• Pre-trained Model: AraBERT.
• Dataset: Custom Arabic First Aid Dataset.

📁 Repository Structure

• Code/: Contains Copy_of_ML_M3.ipynb (Full training and evaluation pipeline).
• Dataset/: The structured combined_first_aid_data.csv file.
• Report/: Detailed technical documentation (PDF).

👥 The Team

Developed by Computer Science students at King Faisal University:

• Atekah Hussain Aljafar
• Anfal Ahmad
• Hanan Alsafran
• Ferdos Kamal
• Supervised by: Prof. Alaa Sagheer

---

Part of the Machine Learning Course (CS-415) - KFU.

Connect with me on LinkedIn for more projects! https://www.linkedin.com/in/ateka-hussain/