🔥 Fire Propagation Stage 1 — Fire Occurrence Prediction

This project builds a physics-informed machine learning model to detect and predict the occurrence of forest fires, using physical fire indices such as ISI, FFMC, BUI, and FWI. The approach combines:

• 🔍 Correlation filtering of physical variables
• 🧠 Unsupervised clustering into positive and negative fire signal dimensions
• 🧠 Neural network modeling using PyTorch
• ✅ High accuracy on structured fire datasets (e.g., Algeria)

📁 Contents

• Fire Propagation Stage 1 - Fire Occurence.ipynb
  Main notebook containing full pipeline from raw data to model evaluation.

• Required Data File: Cleaned_Algerian_Forest_Fire_Data.csv (available in the repository, it is a cleaned version of 'algerian+forest+fires+dataset' from UCI Machine Learning Repository)

⚙️ Instructions

1. Clone this repository or download the .ipynb notebook.
2. Place Cleaned_Algerian_Forest_Fire_Data.csv in the same folder as the notebook.
3. ⚠️ Update the file path in the notebook if needed:

   pd.read_csv("Cleaned_Algerian_Forest_Fire_Data.csv")  # Change this if your file path differs

4. Run the notebook end-to-end in Jupyter or JupyterLab.

📊 Model Highlights

• Applies cluster-based latent inputs
• Learns propagation likelihood from structured correlations
• Recall on fire occurrence: 1.00 (100%)
  Accuracy: up to 96% on Algerian data; ~55% on UCI

📄 License

MIT License — you are free to reuse, extend, or build upon this work.

---

Scientific Note:
This stage focuses only on binary fire occurrence. Future stages may expand to fire spread modeling, rate prediction, or geographic simulation.---

📦 Dataset Acknowledgment

This study is based on the Algerian Forest Fires Dataset
📥 Available via UCI Repository

The dataset includes meteorological and fire data from two regions in Algeria (Sidi-Bel Abbes and Bejaia) between June and September 2012.
➤ Cleaned_Algerian_Forest_Fire_Data.csv

---

📖 Citation

If you use this repository or model in academic work:

@misc{dewmini2025firestage1,
  title={Physics-Informed Fire Prediction – Stage 1: Fire Occurrence},
  author={Dewmini Gunasekera},
  year={2025},
  note={GitHub repository},
  url={https://github.com/dewminigunasekera/physics-informed-fire-prediction-occurrence}
}