# State Farm Distracted Driver Detection Model

## Model Description

This is a YOLOv8 classification model trained on the State Farm Distracted Driver Detection dataset. The model is designed to classify driver behavior into 10 distinct categories to detect distracted driving activities.

## Model Architecture

- **Model**: YOLOv8n-cls (YOLOv8 Nano for classification)
- **Input Size**: 224x224 pixels
- **Number of Classes**: 10
- **Parameters**: 1.45 million
- **Training Framework**: Ultralytics YOLO

## Class Labels

The model predicts the following 10 classes of driver behavior:

- c0: safe driving
- c1: texting - right
- c2: talking on the phone - right
- c3: texting - left
- c4: talking on the phone - left
- c5: operating the radio
- c6: drinking
- c7: reaching behind
- c8: hair and makeup
- c9: talking to passenger

## Training Details

- **Training Epochs**: 12
- **Final Training Loss**: 0.0603
- **Validation Accuracy**: 99.6%
- **Top-5 Accuracy**: 100%
- **Training Time**: 0.27 hours
- **Dataset Split**: 70% train, 15% validation, 15% test
- **Total Training Images**: 15,692
- **Total Validation Images**: 3,358

## Performance Metrics

The model achieved excellent performance on the validation set:

- Top-1 Accuracy: 99.6%
- Top-5 Accuracy: 100%
- Consistent convergence across all epochs
- Low validation loss indicating good generalization

## Installation and Usage

### Requirements

```bash
pip install ultralytics
```

## Basic Usage

```
from ultralytics import YOLO

# Load the model from Hugging Face
model = YOLO('Safe-Drive-TN/State-farm-detection')

# Perform prediction on an image
results = model.predict('path_to_your_image.jpg')

# Get prediction results
predicted_class = results[0].probs.top1
confidence_score = results[0].probs.top1conf
class_names = results[0].names

print(f"Predicted class: {class_names[predicted_class]}")
print(f"Confidence: {confidence_score:.2%}")
````


## Dataset Information

The model was trained on the State Farm Distracted Driver Detection dataset, which contains images of drivers performing various activities in a vehicle. The dataset ensures driver separation between train and test splits to prevent data leakage.

## Limitations

The model was trained in a controlled environment and may not generalize perfectly to real-world driving conditions

Performance may vary with different camera angles, lighting conditions, and vehicle types

The model classifies single frames and does not incorporate temporal information

## Citation

If you use this model, please cite:

```
@misc{state-farm-detection-2025,
  title = {State Farm Detection in Vehicle — Yolov8n-cls},
  author = {Malek Messaoudi, Yassine Mhirsi},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Safe-Drive-TN/State-farm-detection}}
}

````


## License

License : MIT