🌾 YIELDICT AI — Smart Crop Yield Prediction & Farm Decision Support

YIELDICT AI is a machine learning agricultural intelligence and decision-support platform designed to forecast crop yields, evaluate farm economics, and provide risk-mitigated crop recommendations for Indian agriculture.

Powered by an ensemble Random Forest Regression Pipeline ($R^2 = 0.9805$), the platform transforms raw agro-climatic inputs into monetary insights, optimal fertilizer application curves, and climate drought simulations.

---

📸 Key Features

1. 🌾 High-Precision Yield & Revenue Forecasting

• Multi-Parameter Inference: Evaluates 55 crop types across 30 Indian states and 6 agricultural seasons.
• Synchronized Controls: Interactive numeric boxes and sliders for Cultivated Area, Annual Rainfall, Fertilizer Usage, and Pesticide Application.
• Monetary Valuation: Calculates total farm output ($Yield \times Area$) and expected gross revenue in INR (₹ Lakhs / ₹ Crores) mapped to Government Minimum Support Price (MSP) benchmarks.

2. 🏆 Multi-Crop Profitability Leaderboard

• Evaluates alternative crops viable in the selected state and season.
• Ranks top alternative opportunities and displays an interactive Altair Revenue Bar Chart to maximize farm returns.

3. 🧪 Fertilizer ROI & Profit Sweet-Spot Optimizer

• Simulates marginal fertilizer cost (₹30/kg) vs. crop yield returns across $25% - 200%$ dosage variations.
• Pinpoints the Economic Sweet-Spot Dosage (kg/ha) and calculates potential net profit uplift.
• Embeds an interactive Fertilizer Dosage vs. Net Profit Curve.

4. 🌧️ Climate Stress & Drought Resilience Simulator

• Stress-tests farm yields against 5 monsoon rainfall scenarios ($-50%$ drought deficit to $+50%$ excess monsoon).
• Computes a Drought Resilience Score (%) and categorizes the farm under automated climate risk tiers (🛡️ Low Risk / ⚠️ Moderate Risk / 🚨 High Risk).

5. 🎨 Botanical Paper & Nature Glassmorphism UI

• Curated emerald-forest palette (#2d6a4f, #52b788, #1b382b), frosted glass surfaces, and responsive typography (Playfair Display + Plus Jakarta Sans).

---

🛠️ Tech Stack & Dependencies

Layer	Technology
Frontend UI	Streamlit, HTML5, Vanilla CSS3 (Custom Botanical Paper Theme)
Data Visualization	Altair Interactive Visualizations
Machine Learning	Scikit-Learn (Pipeline, ColumnTransformer, OneHotEncoder, RandomForestRegressor)
Data Manipulation	Pandas, NumPy
Serialization	Joblib, Git LFS

---

🚀 Step-by-Step Setup Guide

1. Clone the Repository

git clone https://github.com/TeamKinetix/Yieldict_AI.git
cd Yieldict_AI

2. Initialize Git LFS (for Model Weights)

The trained model (crop_yield_model.pkl, 175.5 MB) is tracked with Git Large File Storage (LFS):

git lfs install
git lfs pull

3. Set Up a Virtual Environment

# Windows
python -m venv venv
.\venv\Scripts\activate

# macOS / Linux
python3 -m venv venv
source venv/bin/activate

4. Install Dependencies

pip install -r requirements.txt

---

💻 How to Run

Method 1: Launch the Main Web Application (Recommended)

streamlit run "ML base_line trainning/ML base_line trainning/app.py"

Or use the launcher wrapper:

streamlit run "ML base_line trainning/app.py"

Once started, open http://localhost:8501 in your browser.

---

Method 2: Launch the Standalone Optimizer API Server

For microservice setups or programmatic querying:

python optimizer/server.py

Access the advisory endpoint at http://localhost:8000.

---

Method 3: Run the CLI Decision Advisory Tool

For fast command-line farm simulations:

python optimizer/crop_optimizer.py

---

🧠 Model Training & Pipeline Architecture

The model is trained on 19,689 validated historical agricultural records using a single scikit-learn Pipeline:

Input Features (7)
│
├── Categorical Features ['crop', 'season', 'state']
│   └── OneHotEncoder(handle_unknown='ignore', sparse_output=False) -> [88 dimensions]
│
├── Numerical Features ['area', 'annual_rainfall', 'fertilizer', 'pesticide']
│   └── Passthrough -> [7 dimensions]
│
└── Regressor
    └── RandomForestRegressor(n_estimators=100, max_depth=None, random_state=42, n_jobs=-1)

Model Performance Metrics

• Coefficient of Determination ($R^2$): 0.9805 (98.05% variance explained)
• Mean Absolute Error (MAE): 8.79 t/ha
• Root Mean Squared Error (RMSE): 118.42 t/ha

How to Retrain the Model:

To reproduce or retrain the pipeline from scratch, open and run all cells in:

important code/Baseline_Model.ipynb

The notebook automatically outputs the updated crop_yield_model.pkl pipeline.

---

📂 Project Directory Structure

CropYield/
│
├── .gitattributes                  # Git LFS configuration (*.pkl tracking)
├── .gitignore                      # Tailored repository ignore rules
├── .streamlit/                     # Streamlit theme and viewport configuration
│   └── config.toml
│
├── 6 files/                        # PRD, Architecture, and Design specifications
│   ├── PRD.md
│   ├── architecture.md
│   ├── design.md
│   └── phases.md
│
├── important code/                 # Training notebooks & model checkpoints
│   ├── Baseline_Model.ipynb        # Model training & evaluation notebook
│   └── crop_yield_cleaned.csv      # Agricultural dataset (19,689 records)
│
├── ML base_line trainning/         # Web Application core
│   ├── ML base_line trainning/
│   │   ├── app.py                  # Main Streamlit web application
│   │   ├── crop_yield_model.pkl    # Serialized scikit-learn Pipeline (175.5 MB)
│   │   ├── crop_yield_cleaned.csv  # Cleaned dataset
│   │   └── background.png          # UI graphics & branding
│   └── app.py                      # Root launcher script
│
├── optimizer/                      # Advisory & Simulation Engine
│   ├── __init__.py                 # Package initializer
│   ├── crop_optimizer.py           # Multi-crop, fertilizer ROI & climate stress logic
│   ├── msp_data.py                 # Crop pricing, MSP data & categories
│   └── server.py                   # Standalone advisory service (Port 8000)
│
├── processed data for training/    # Cleaned datasets & EDA documentation
│   ├── STEP_3_EDA_REPORT.md
│   └── crop_yield_cleaned.csv
│
├── requirements.txt                # Python package dependencies
└── README.md                       # Master project documentation

---

👥 Authors & Team

Developed by Kinetix. For feedback, inquiries, or contributions, please open a GitHub Issue or Pull Request.