🌱 SoilSense: Explainable Soil Degradation & Health Analysis using Machine Learning

---

🌟 Overview

SoilSense predicts soil health and degradation using a hybrid ML model (XGBoost + Random Forest) with Optuna hyperparameter optimization.

It calculates the Soil Degradation Index (SDI) and classifies soil as:

• 🟢 Healthy
• 🟡 Low
• 🟠 Moderate
• 🔴 Highly Degraded

Using SHAP explainability, SoilSense provides actionable recommendations for farmers on which soil parameters to adjust to improve soil health. Users can download detailed reports.

💡 Real-Time Insights: The dashboard updates predictions and recommendations in real-time when input parameters are adjusted, allowing farmers to instantly see the impact of changes on soil health.

---

🚀 Features

• 🔹 Hybrid ML Model: XGBoost + Random Forest
• 🔹 Optuna-based hyperparameter tuning
• 🔹 Predicts Soil Degradation Index (SDI) and class
• 🔹 SHAP explainability with actionable recommendations
• 🔹 Interactive Streamlit dashboard with real-time updates
• 🔹 Report generation and download
• 🔹 Synthetic dataset creation for training and evaluation

---

💡 Novelty Points

• 🌱 Hybrid Model Approach: Combines XGBoost and Random Forest for improved predictive performance and robustness.
• 🎯 Optuna Optimization: Automatically tunes hyperparameters for optimal model performance.
• 🔍 Explainability with SHAP: Provides farmers with actionable insights on which soil parameters to adjust.
• 📊 Integrated Regression & Classification: Predicts both Soil Degradation Index and discrete soil health classes.
• 💾 Synthetic Dataset Generation: Allows training and testing in the absence of extensive real-world soil datasets.
• 🖥️ Interactive Dashboard: User-friendly interface for real-time predictions and report downloads.
• ⚡ Real-Time Parameter Adjustment: Farmers can modify input parameters and immediately see changes in soil health predictions.

---

📝 Input Features

• 🌾 N: Nitrogen content
• 🌾 P: Phosphorus content
• 🌾 K: Potassium content
• 💧 Moisture: Soil moisture level
• 🌡️ Temperature: Ambient temperature
• 🌬️ Humidity: Ambient humidity

---

🤖 Machine Learning Approach

• 🌲 Random Forest: Captures non-linear relationships
• ⚡ XGBoost: Gradient boosting for high performance
• 🔗 Hybrid Model: Combines both for better accuracy
• 🎯 Optuna Optimization: Automatic hyperparameter tuning
• 🔍 SHAP Explainer: Provides interpretability and actionable recommendations

---

📊 Dataset

• File: soil_dataset.csv
• Synthetic dataset created for realistic soil health simulations

---

💾 Model Files

📁 File	📄 Description
rf_reg.pkl	Random Forest regression model
xgb_reg.pkl	XGBoost regression model
rf_cls.pkl	Random Forest classification model
xgb_cls.pkl	XGBoost classification model
label_encoder.pkl	Label encoder for soil classes

---

📈 Model Performance

🧮 Regression Metrics

Metric	Train	Test
RMSE	0.7800	1.5766
R²	0.9990	0.9960

🏷️ Classification Metrics

Metric	Train	Test
Accuracy	99.94%	96.25%
F1 Score	0.9994	0.9622

🧩 Confusion Matrices

Train Confusion Matrix:

Actual \ Pred	🟢 Healthy	🟡 Low	🟠 Moderate	🔴 High
🟢 Healthy	1614	0	0	0
🟡 Low	4	1602	0	0
🟠 Moderate	0	0	1588	0
🔴 High	0	0	0	1592

Test Confusion Matrix:

Actual \ Pred	🟢 Healthy	🟡 Low	🟠 Moderate	🔴 High
🟢 Healthy	386	0	0	0
🟡 Low	12	373	9	0
🟠 Moderate	0	30	381	9
🔴 High	0	0	0	400

---

💻 Installation

git clone https://github.com/<your-username>/SoilSense.git
cd SoilSense
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
python dataset.py
python model.py
streamlit run app.py

---

🛠️ Usage

1. Upload your soil dataset (soil_dataset.csv) with the input features.
2. Predict the Soil Degradation Index and soil class.
3. View SHAP explanations and actionable recommendations.
4. Adjust input parameters to see real-time changes in soil health predictions.
5. Download a detailed report.

---

📂 Folder Structure

SoilSense/
│
├── app.py                # Streamlit dashboard
├── dataset.py            # Dataset creation and preprocessing
├── model.py              # Model training and prediction functions
├── rf_cls.pkl            # Random Forest classification model
├── xgb_cls.pkl           # XGBoost classification model
├── rf_reg.pkl            # Random Forest regression model
├── xgb_reg.pkl           # XGBoost regression model
├── label_encoder.pkl     # Label encoder for soil classes
├── soil_dataset.csv      # Synthetic soil dataset
├── requirements.txt
└── soil_env/             # Virtual environment 

---

🔮 Future Enhancements

• 🌐 Integration with IoT soil sensors for real-time monitoring
• 🛰️ Incorporation of satellite imagery for spatial soil health analysis
• 📱 Deployment as a web app for farmers and agronomists

---