YasmineCupcake/DataMining_Project

A datamining project for predicting fire incidents in Algeria + Tunisia # DATA MINING – Practical Work Project **Forest Fire Prediction and Risk Analysis** ## Overview Forest fires represent a major environmental and socio-economic challenge, causing vegetation loss, soil degradation, and severe ecological damage. Early prediction of fire occurrence is essential for effective prevention and management. This project develops a **data-driven predictive system** using **soil characteristics** and **climate variables** to forecast forest fire occurrence and identify high-risk zones. Both **supervised** and **unsupervised** machine learning techniques are applied, with implementations developed **from scratch** and compared with Scikit-learn models. ## Study Area - **Countries**: Algeria & Tunisia (grouped in the same dataset) - **Year**: 2024 - **Academic Context**: Students: ALLAF Chaima + Boutkedjirt Aya USTHB – Faculty of Computer Science IASD – M2 SII Academic Year: 2025/2026 --- ## Project Objectives - Collect and preprocess soil and climate data relevant to fire prediction - Predict fire occurrence using supervised learning algorithms - Identify natural clusters and high-risk fire areas using unsupervised learning - Evaluate models using standard performance metrics - Provide interpretable insights for fire risk analysis --- ## Data Mining Methodology ### Step 1: Data Analysis and Preprocessing - Exploratory Data Analysis (EDA) - Data cleaning and preprocessing - Data integration (soil, climate, fire, elevation) - Feature engineering ### Step 2: Supervised Machine Learning Algorithms implemented **from scratch**: - K-Nearest Neighbors (KNN) - Decision Trees - Random Forest Evaluation: - Accuracy - Precision - Recall - F1-score - ROC-AUC Comparative analysis with **Scikit-learn implementations**. ### Step 3: Unsupervised Machine Learning (Clustering) Algorithms implemented **from scratch**: - K-Means - DBSCAN - CLARANS Evaluation and comparison with Sc …