Algerian Forest Fire Prediction Project

Project Overview

Objective:
Develop a predictive model to forecast forest fires in Algeria using historical weather and fire data. The project focuses on two regions – Bejaia (Northeast) and Sidi Bel-Abbes (Northwest) – with data collected from June 2012 to September 2012.

Key Achievements:

• Achieved 98% accuracy in predicting forest fire occurrences.
• Deployed a Flask-based web application for real-time predictions.
• Utilized multiple regression techniques for robust model performance.

---

Technologies & Tools

• Programming Language: Python
• Data Processing: Pandas, NumPy
• Data Visualization: Matplotlib, Seaborn
• Machine Learning Models: Linear Regression, Ridge, Lasso, ElasticNet (using Scikit-learn)
• Web Development: Flask
• Model Serialization: Pickle

---

Dataset Overview

• Total Instances: 244
  • Fire: 138 instances
  • Not Fire: 106 instances
• Data Collection Period: June 2012 – September 2012
• Key Attributes:
  • Date: Day, month, and year information.
  • Temp: Temperature at noon (22°C to 42°C).
  • RH: Relative Humidity (21% to 90%).
  • Ws: Wind Speed (6 km/h to 29 km/h).
  • Rain: Total rainfall in mm (0 to 16.8 mm).
  • FWI Components: Includes metrics like Fine Fuel Moisture, Duff Moisture, Drought Code, etc.

---

Methodology & Workflow

1. Data Cleaning & Preprocessing

   • Handled missing values and standardized features.
   • Transformed raw data for effective modeling.

2. Exploratory Data Analysis (EDA)

   • Visualized attribute distributions and correlations using Seaborn and Matplotlib.
   • Identified key predictors affecting fire occurrences.

3. Feature Engineering & Selection

   • Extracted and selected features to enhance model performance.
   • Applied statistical techniques to validate feature importance.

4. Model Training & Evaluation

   • Implemented and compared several regression models:
     • Linear Regression
     • Ridge Regression
     • Lasso Regression
     • ElasticNet Regression
   • Evaluated models using metrics like Mean Squared Error (MSE) and R².

5. Model Serialization

   • Serialized the best performing model with Pickle for deployment.

6. Web Application Development

   • Developed a user-friendly Flask web app to allow users to input data and receive real-time predictions.
   • Integrated interactive data visualizations within the web interface.

7. Deployment

   • Deployed the Flask application on a cloud platform for accessibility.
   • Ensured scalability and responsiveness in the production environment.

---

Visual Demonstrations

Web Application Interface

Model Prediction Dashboard

---

Results & Impact

• Accuracy: 98% in predicting forest fires.
• Real-Time Predictions: Enabled dynamic forecasting via the deployed web app.
• Business Impact: Provides a tool for early warning systems, potentially reducing forest fire damage and aiding in resource allocation.

---

How to Get Started

1. Clone the Repository:

   git clone https://github.com/Ganglet/Algerian-Forest-Fire-Model.git