BetterCallEkangsh/Algerian-Forest-Fire-FWI-Prediction

This project analyzes the Algerian Forest Fires dataset — 244 weather observations from two regions of Algeria (Bejaia in the northeast, Sidi Bel-Abbes in the northwest), collected over June–September 2012 — to predict the Fire Weather Index (FWI), a composite score used by fire-danger rating systems worldwide. # 🔥 Algerian Forest Fire — FWI Prediction ### Regression modeling of the Fire Weather Index using regularized linear models --- ## 📌 Overview This project analyzes the **Algerian Forest Fires dataset** — 244 weather observations from two regions of Algeria (Bejaia in the northeast, Sidi Bel-Abbes in the northwest), collected over June–September 2012 — to predict the **Fire Weather Index (FWI)**, a composite score used by fire-danger rating systems worldwide. The work is split into two notebooks: | Notebook | Purpose | |---|---| | `Algerian_Forest_Fire_Data_Analysis.ipynb` | Cleaning, wrangling, and exploratory data analysis | | `Algerian_Forest_Model_Training.ipynb` | Feature selection, scaling, and regression modeling | --- ## 🌍 Dataset - **244 instances** — 122 per region, merged from two raw CSV blocks - **11 weather/FWI-system features** + 1 target class label - Region split encoded as a binary feature (`0` = Bejaia, `1` = Sidi Bel-Abbes) - Class balance: **138 fire** vs **106 not-fire** days | Feature | Description | Range | |---|---|---| | Temperature | Noon temperature (°C) | 22 – 42 | | RH | Relative Humidity (%) | 21 – 90 | | Ws | Wind speed (km/h) | 6 – 29 | | Rain | Total rainfall (mm) | 0 – 16.8 | | FFMC | Fine Fuel Moisture Code | 28.6 – 92.5 | | DMC | Duff Moisture Code | 1.1 – 65.9 | | DC | Drought Code | 7 – 220.4 | | ISI | Initial Spread Index | 0 – 18.5 | | BUI | Buildup Index | 1.1 – 68 | | **FWI** | **Fire Weather Index (target)** | 0 – 31.1 | | Classes | `fire` / `not fire` | — | --- ## 🧹 Data Cleaning - Merged the two region-specific blocks of the raw CSV and stamped a `Region` column - Dropped a malformed header row separating the two regions, and rows with nulls - Stripped whitespace from column names and category labels (`" fire"` → `"fire"`) - Cast day/month/year and integer-valued weather fields to `int`, remaining FWI-system columns to `float` - Persisted a clean, analysis-ready CSV for the modeling notebook ## 📊 Expl …