Adams-Nafisah/ML-BASED-DECISION-SUPPORT-SYSTEM-FOR-MAIZE-YIELD-PREDICTION-IN-NIGERIA

This projects uses historical Nigerian climate and soil data to train different machine learning models and build a stacked ensemble model for optimal performance of maize yield prediction. Maize Yield Prediction System with Explainable AI Project Summary This project develops an end-to-end machine learning system for predicting maize yield using environmental and soil data. The pipeline includes data analysis, feature engineering, model training, and explainability using SHAP. Initial models showed near-perfect accuracy but were found to rely on geographic identifiers, leading to overfitting. A refined scientific model was developed using only agronomically meaningful variables, resulting in more realistic performance (R² ≈ 0.84). The final system balances predictive accuracy with interpretability and real-world applicability. A Streamlit application is included to provide an interactive interface where users can input environmental conditions and receive yield predictions, explanations, and actionable agricultural recommendations. project/ │ ├── data/ │ └── final-data-Before FS.csv │ ├── models/ │ ├── maize_yield_ensemble.pkl │ ├── scientific_rf.pkl │ ├── scientific_xgb.pkl │ ├── scientific_lgbm.pkl │ ├── scaler.pkl │ └── scientific_features.json │ ├── notebooks/ │ └── main_analysis.ipynb │ ├── figures/ │ ├── eda_yield_distribution.png │ ├── eda_scatter_updated.png │ ├── shap_scientific_summary.png │ └── prediction_accuracy_analysis.png ├── app/ │ └── app.py │ │ └── README.md Setup Instructions 1. Create Environment python -m venv venv source venv/bin/activate # Mac/Linux venv\Scripts\activate # Windows 2. Install Dependencies pip install pandas numpy matplotlib seaborn scikit-learn xgboost lightgbm shap joblib streamlit 3. Optional (for notebook usage) pip install jupyter How to Run the Project Step 1: Load Dataset Ensure the dataset file is placed in the data/ folder: data/final-data-Before FS.csv Step 2: Run Notebook Open and execute: notebooks/01_maize_yield_modeling.ipynb This will: Load and clean data Perform exploratory analysis Engineer features Train models Evaluate perf …