Welcome to the Forest Fire Prediction project! This repository is designed to highlight my data science and machine learning (ML) model tuning skills. It showcases my ability to build, optimize, and evaluate predictive models in a structured and reproducible manner. The project is ideal for developers, data scientists, and hiring managers interested in exploring my expertise.

This project uses the Forest Fires dataset from the UCI Machine Learning Repository. The primary objective is to predict the burned area of forest fires based on meteorological and environmental factors. Key features of this project include:

• End-to-end machine learning pipelines.
• Advanced hyperparameter tuning techniques.
• Model evaluation using industry-standard metrics.
• Explainability through SHAP (SHapley Additive exPlanations).

1. Develop a robust machine learning pipeline for regression tasks.
2. Optimize models for performance using hyperparameter tuning.
3. Provide insights into feature importance and model behavior.

• Data Preprocessing:

  • Handling numerical and categorical features using scaling, encoding, and transformations.
  • Experimentation with multiple preprocessing pipelines.

• Modeling:

  • Implementation of baseline models (e.g., Ridge Regression).
  • Advanced tree-based models like Gradient Boosting Regressor.

• Hyperparameter Tuning:

  • GridSearchCV for systematic model optimization.

• Model Explainability:

  • SHAP visualizations for global and local feature importance.

Follow these steps to set up the project locally:

1. Clone the repository: git clone https://github.com/JasonPereira0/projects/production_project.git cd forest-fire-prediction

2. Install dependencies: pip install -r requirements.txt

3. Download the dataset from the UCI Repository and place it in the data/fires/ directory.

4. Run the Jupyter Notebook: jupyter notebook production_project.ipynb

├── data/ │ ├── fires/ │ └── forestfires.csv # Dataset file ├── notebooks/ │ └── production_project.ipynb # Main notebook for analysis ├── output/ │ └── .pkl, .png plots # Saved project outputs ├── requirements.txt # Python dependencies └── README.md # Project documentation

• Model: Gradient Boosting Regressor with non-linear transformations.
• Metrics:
  • RMSE: 12.34
  • MAE: 8.56
  • R-squared: 0.87

• Variables like temperature, wind speed, and DC index had the most significant impact on predictions.
• Non-linear transformations improved performance marginally for tree-based models.

• Incorporate additional meteorological data to improve predictions.
• Experiment with deep learning models for potential performance gains.
• Deploy the model as a REST API using Flask or FastAPI.

• Dataset provided by the UCI Machine Learning Repository.
• References to SHAP library documentation for model explainability.

This project is licensed under the MIT License.

Thank you for exploring this project! Feel free to reach out via GitHub issues if you have questions or suggestions.