namimaleki/Algerian-Forest-Fire-Predictor

# Algerian Forest Fire Predictor ## About This Project This was one of my first end-to-end machine learning projects, built as part of a Udemy ML course I took to get hands-on experience with the full ML pipeline. The idea was to work through a real dataset, understand what the data is telling me, clean it properly, experiment with different models, and eventually build a web app that could take inputs and return a prediction. I rebuilt and documented everything myself so I could actually understand each step rather than just running someone else's code. --- ## The Problem The dataset comes from two regions in Algeria — Bejaia and Sidi-Bel-abbes — and covers weather observations recorded daily from June to September 2012. Each row includes temperature, humidity, wind speed, rain, and several fire weather index components. The goal is to predict the **Fire Weather Index (FWI)** — a continuous value that represents how severe the fire weather conditions are on a given day. Higher FWI = more dangerous conditions for fire spread. This makes it a regression problem (continous target variable). --- ## My Approach **1. EDA and Data Cleaning** The raw dataset had a few issues — the two regions were stacked in one file with a separator row in the middle, some columns had whitespace in their names, and the data types were all wrong because of the formatting. I worked through all of that, added a Region column to preserve that information, and saved a clean version for modeling. From there I explored the data visually — distributions, class balance, and correlations — to understand what I was working with before touching any models. **2. Feature Selection** I used a correlation threshold to identify and drop features that were too similar to each other (BUI and DC both correlated above 0.85 with other features). Keeping them would have introduced multicollinearity and made the regularized models less stable. **3. Feature Scaling** Applied StandardScaler to normali …