Algerian Forest Fire Weather Index Predictor

An educational regression project using the Algerian Forest Fires dataset. A Flask form applies a saved scaler and Ridge model to estimate the Fire Weather Index (FWI).

Run locally

git clone https://github.com/AtulJ505/Forest_Fire.git
cd Forest_Fire
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python application.py

Open http://127.0.0.1:5000. On Windows, activate with .venv\Scripts\activate. The application loads models/ridge.pkl and models/scaler.pkl relative to its own directory. Only load trusted pickle artifacts; use scikit-learn versions compatible with the saved models or retrain them.

Inputs and output

The form accepts Temperature, relative humidity (RH), wind speed (Ws), Rain, FFMC, DMC, ISI, Classes and Region in that order. Consult the notebook for the dataset's units and categorical encoding. Missing, malformed, NaN and infinite values return HTTP 400 before running a model. A successful request displays the predicted FWI.

Tests

pip install Flask pytest
python -m pytest -q

The route tests use mocked model outputs. They verify form rendering, input validation, feature ordering, and artifact-path resolution without retraining or claiming a model-accuracy result. CI runs these tests on Python 3.11.

Scope and limitations

This repository demonstrates data preprocessing, regression and a Flask prediction interface. Model quality must be evaluated on held-out data with documented metrics and dependency versions. It is not a validated fire-warning service or an operational safety tool. A live deployment URL is not currently documented.

Contributing

Include a minimal reproduction for a bug and a regression test when changing application behavior. Keep dataset attribution, preprocessing details and evaluation methodology visible when updating the model.