harsh2kum/Fire-Weather-Index-Prediction-ML

End-to-end machine learning project on the Algerian Forest Fires dataset, including EDA, feature engineering, linear regression modeling, and deployment using Flask.,AWS # 🔥 Fire Weather Index (FWI) Prediction | Machine Learning Project An **end-to-end Machine Learning project** built using the **Algerian Forest Fires dataset** to predict the **Fire Weather Index (FWI)**. This project demonstrates the complete ML lifecycle — from **EDA and feature engineering** to **model deployment using Flask**. --- ## 📌 Project Overview Forest fires are strongly influenced by weather conditions such as temperature, humidity, wind speed, and rainfall. The **Fire Weather Index (FWI)** is an important indicator used to estimate fire risk intensity. In this project, I: - Analyzed the dataset using Exploratory Data Analysis (EDA) - Performed feature engineering and scaling - Trained a regression model using **Ridge Regression** - Deployed the trained model as a **Flask web application** --- ## 📊 Dataset Information - **Dataset Name:** Algerian Forest Fires Dataset - **Source:** UCI Machine Learning Repository - **Regions Covered:** Bejaia & Sidi Bel-Abbes (Algeria) - **Target Variable:** Fire Weather Index (FWI) ### Key Features - Temperature - Relative Humidity (RH) - Wind Speed (Ws) - Rain - FFMC - DMC - ISI - Classes - Region --- ## 🧠 Machine Learning Pipeline 1. **Exploratory Data Analysis (EDA)** - Distribution analysis - Correlation study - Fire vs non-fire patterns 2. **Feature Engineering** - Data cleaning - Encoding categorical features - Feature scaling using `StandardScaler` 3. **Model Training** - Algorithm: **Linear Regression with Ridge Regularization** - Helps prevent overfitting 4. **Model Evaluation** - R² Score - Mean Absolute Error (MAE) - Mean Squared Error (MSE) 5. **Deployment** - Model and scaler saved using `pickle` - Flask-based web application for real-time predictions --- ## 🛠️ Tech Stack - **Programming Language:** Python - **Libraries:** NumPy, Pandas, Matplotlib, Seaborn, Scikit-learn - …