Skip to content

Predictive Modelling ‐ Burnt Area Prediction Model Guide

Jump to bottom
Eden Oliel edited this page Apr 8, 2025 · 1 revision

Model Overview

This guide documents the development of a machine learning model to forecast wildfire behavior. The model predicts the burned area of a wildfire based on meteorological and environmental data. It integrates with our existing wildfire visualization platform via a RESTful API.

The current model is trained on the Forest Fires Dataset, which is geographically limited to Montesinho Natural Park in northeast Portugal. While suitable for prototyping, results may not generalize to other regions without retraining on localized data.

Dataset Selection

Final Dataset : Forest Fires Dataset (UCI Repository)

  • Well-structured and includes relevant features
  • Geographically limited to Portugal, but suitable for prototyping

Model Architecture

Model Type : Support Vector Regression (SVR)

  • Kernel: Radial Basis Function (RBF)
  • Chosen for its ability to model non-linear relationships

Run Instructions

Follow run instructions in Developer Guide in README.md

ML Model Development

cd ml
python3 -m venv myenv
source myenv/bin/activate
pip install -r requirements.txt
jupyter notebook

Running the API

cd ml-api
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload

For more background on model selection, dataset considerations, and design decisions, see the related GitHub issue: Machine Learning Predictive Model - Burnt Area Prediction

Clone this wiki locally