This project predicts the Fire Weather Index (FWI), a key indicator of wildfire risk, using meteorological data. The dataset consists of weather observations recorded from June to September 2012 for 2 regions ,namely the Bejaia region located in the northeast of Algeria and the Sidi Bel-abbes region located in the northwest of Algeria.. Three linear regression models (Linear, Ridge, and Lasso) were applied, with hyperparameter tuning to improve performance.
You can access the live app here
The dataset includes the following features:
- Temperature (°C): 22 - 42
- Relative Humidity (%): 21 - 90
- Wind Speed (km/h): 6 - 29
- Rainfall (mm): 0 - 16.8
- Fine Fuel Moisture Code (FFMC): 28.6 - 92.5
- Duff Moisture Code (DMC): 1.1 - 65.9
- Drought Code (DC): 7 - 220.4
- Initial Spread Index (ISI): 0 - 18.5
- Buildup Index (BUI): 1.1 - 68
- Fire Weather Index (FWI) (Target Variable): 0 - 31.1
- Install Dependencies: pip install -r requirements.txt
- Clone the repository:
- git clone https://github.com/abhinavbatra06/fire-index-prediction.git
- cd fire-weather-index-prediction
- jupyter notebook
- Data Cleaning
- Removed null values and corrected column names.
- Standardized categorical labels and converted region info into a binary column (is_sidi_bel_region)
- Added is_august as a seasonality feature (August had the highest fire occurrences).
- Removed highly correlated features (BUI & DC) to reduce multicollinearity.
- Train-Test Split: 80-20 split.
- Regression Models Used:
- Linear Regression (Baseline)
- Ridge Regression (Best alpha = 5, L2 regularization)
- Lasso Regression (Best alpha = 0.01, L1 regularization)
Model | R² Score | Mean Squared Error (MSE)
Linear Regression | 0.81 | 2.43
Ridge Regression | 0.81 | 2.40
Lasso Regression | 0.81 | 2.40
- Minimal improvement in Ridge & Lasso over Linear Regression.
- Fire risk is highly seasonal, with August showing the highest fire occurrences.
- The highest values of coefficients are for - Initial Spread Index (ISI) & Duff Moisture Code (DMC)