cibu-eduard-30122/AgriVuln-A

I am building an AI-powered system that fuses multi-sensor satellite data with socioeconomic indicators to map, predict, and visualize agricultural and social vulnerability at high resolution for Kenya. AgriVulnAI is an end-to-end geospatial machine learning system developed to estimate agricultural vulnerability in Kenya by integrating: Sentinel-2 vegetation indices (NDVI, EVI) CHIRPS precipitation PM2.5 air quality CAMS aerosol indicators Hydrological layers (Water Occurrence) Socio-economic data (WorldPop 2020) A LightGBM classifier trained on vulnerability ground truth The project produces: A composite Vulnerability Index v2 A 4-level vulnerability clustering (Low → Very High) An interactive geospatial dashboard for exploration Feature importance explanations via SHAP The pipeline is optimized for research, policy-making, and early-stage climate-risk product development. Project Structure: AgriVuln-A/ │ ├── dashboard_app.py # Main Streamlit dashboard ├── compute_vulnerability_index_v2.py ├── data_pipeline.py ├── modeling.py # LightGBM model training ├── preprocess.py # Data cleaning & prep │ ├── data/ │ └── processed/ │ ├── ground_truth_with_predictions.csv │ ├── ground_truth_with_predictions_v2.csv │ ├── features_ground_truth.csv │ └── features_ground_truth_with_extra.csv │ ├── results/ │ ├── metrics.csv │ ├── confusion_matrix.csv │ ├── shap_plots/ │ │ └── shap_summary.png # Add this manually │ └── map_points_predictions.png │ ├── figures/ # Interactive HTML maps ├── requirements.txt └── README.md 1. Machine Learning Results Model used: LightGBM, trained on 4,925 labeled points. ✔ Overall performance Metric Value Accuracy 0.965 F1-macro 0.908 Confusion Matrix: [[ 17 0 10 1 0 0] [ 0 1 0 0 0 0] [ 8 0 857 3 0 2] [ 1 0 9 51 0 0] [ 0 0 0 0 2 0] [ 0 0 0 0 0 21]] Interpretation: Class 4 is the dominant and best-predicted class 0 and 7 show moderate confusion → expected due to overlapping spectral patterns Rare classes (1, 9, 10) achieve perfect or near-perfect precision due to clear si …