nihal00753/Fire_Intensity

A binary classification project that predicts whether a forest fire occurred based on meteorological conditions and Fire Weather Index (FWI) system components from two regions of Algeria. # 🔥 Algerian Forest Fire Prediction ### Binary Classification · Feature Engineering · Ensemble Models --- ## 📌 Overview A binary classification project that predicts whether a **forest fire occurred** based on meteorological conditions and Fire Weather Index (FWI) system components from two regions of Algeria. The project goes beyond a simple model fit — it applies multicollinearity analysis, domain-informed feature engineering, SMOTE-based class balancing, and systematic multi-model comparison with hyperparameter tuning, achieving a best ROC-AUC of **~0.93** on the held-out test set. --- ## 📊 Dataset **Algerian Forest Fires Dataset** — UCI Machine Learning Repository - Weather observations from **Bejaia** and **Sidi Bel-abbes** regions, June–September 2012 - **243 observations**, 11 input features, binary target: `fire` / `not fire` - Features include: Temperature, RH (relative humidity), Wind speed (Ws), Rain, and FWI system components (FFMC, DMC, DC, ISI, BUI, FWI) --- ## ✨ What Was Achieved ### 🔬 Data Processing & EDA - Loaded and cleaned the combined Bejaia + Sidi Bel-abbes dataset, handling region separators and mixed-type columns - Performed full **Exploratory Data Analysis (EDA)** with distribution plots, class balance checks, and correlation heatmaps - Identified and documented the **FWI component hierarchy**: FFMC → ISI → FWI (composite), and DMC + DC → BUI → FWI — a structural redundancy that directly informed feature selection ### ⚙️ Feature Engineering Two domain-informed features were created based on fire science principles: - **`heat_stress`** = `Temperature / RH` — captures the combined drying effect of heat and low humidity, a key ignition risk factor - **`wind_fwi`** = `Ws × FWI` — encodes the interaction between wind speed and the composite fire danger index, reflecting how wind amplifies high-risk conditions ### 🧹 Handling Class Imbalance - Applied **SMOTE (Synthetic Minority Over-sampling Technique)** exclusive …