Rithi123456/algerian-wildfire-weather-model

Wildfire risk (FWI) prediction from weather data, with data leakage correction and cross-validated model comparison. # Algerian Wildfire Risk Prediction — Weather-Only Model Predicting the Fire Weather Index (FWI) for Algerian wildfires using only raw meteorological data — explicitly avoiding leakage from FWI-system-derived features, and comparing linear vs non-linear models with proper cross-validation. ## Why This Project Exists Most public versions of this project report R² scores of ~0.98 by including features like `ISI`, `BUI`, `FFMC`, and `DMC` as model inputs. These aren't independent weather measurements — they're intermediate values in the Canadian Forest Fire Weather Index system, calculated FROM raw weather and used to calculate FWI itself: ``` Temperature, RH, Wind, Rain → FFMC, DMC, DC → ISI, BUI → FWI ``` Including these as model features means the model is largely reconstructing a known formula, not learning genuine predictive patterns. **This version uses only Temperature, Relative Humidity, Wind Speed, Rain, and Region as features** — the actual independent weather inputs. ## Results **Regression (predicting FWI), evaluated via 5-fold cross-validation:** | Model | Mean CV R² | Std | |---|---|---| | Random Forest | **0.626** | 0.212 | | Linear Regression | 0.378 | 0.119 | Random Forest meaningfully outperforms Linear Regression — real evidence of non-linear interactions between weather variables (e.g., temperature's effect likely compounds with low humidity rather than just adding to it). Higher variance in Random Forest's fold scores is likely a consequence of the small dataset (~183 training rows), noted honestly rather than hidden. **Classification (fire / not-fire), evaluated via 5-fold cross-validation:** | Model | Mean CV F1 | |---|---| | Random Forest | 0.862 | | Logistic Regression | 0.859 | Essentially tied — added model complexity doesn't help here, so Logistic Regression is preferred for simplicity and interpretability. **Feature importance (Random Forest, regression task):** Rain (0.40) and Relative Humidity (0.34) dominate, followed by Te …