ryma-tharouma/Data-Mining---Fire-Prediction

Geospatial fire risk assessment in Algeria and Tunisia using ML on climate, soil, land cover, and elevation data, with prediction and risk zoning. # Data Mining Project: Fire Prediction ## Overview This project applies multiple machine learning algorithms to predict forest fires using environmental and geographical data. It includes data preprocessing, exploratory data analysis, and comprehensive model evaluation with various resampling strategies. ## Project Structure ``` ├── code/ # Core utilities and models │ ├── data_loader.py # Data loading utilities │ ├── file_tree.py # File structure utilities │ ├── Metrics/ # Custom evaluation metrics │ └── myModels/ # Custom model implementations │ ├── clarans.py │ ├── dbscan.py │ ├── DecisionTree.py │ ├── kmeans.py │ ├── knn.py │ └── RandomForest.py │ ├── notebooks/ # Jupyter notebooks for analysis │ ├── EDA/ # Exploratory data analysis │ │ └── [climate, elevation, fire, landcover, soil analyses] │ ├── processing_1-3/ # Data preprocessing pipelines │ ├── modeling/ # Model training & evaluation │ │ ├── CLARANS/ │ │ ├── DBSCAN/ │ │ ├── DT/ # Decision Tree │ │ ├── kMeans/ │ │ ├── knn/ # K-Nearest Neighbors │ │ └── RF/ # Random Forest │ └── Merge/ # Data merging operations │ └── models/ # Trained model artifacts ``` ## Dataset & Features The project analyzes multiple feature categories: - **Climate**: Temperature, humidity, wind patterns - **Elevation**: Topographic data - **Fire**: Fire occurrence and characteristics - **Landcover**: Vegetation and land use types - **Soil**: Soil properties and composition ## Methods ### Clustering Algorithms - K-Means - CLARANS - DBSCAN ### Classification Algorithms - Decision Tree - K-Nearest Neighbors (KNN) - Random Forest ### Data Balancing Strategies - SMOTE + Tomek Links - K-Means SMOTE - NearMiss - …