Smart Agriculture Advisory System

An ML-based advisory system that recommends suitable crops for Indian farmers by state and district, land size, and region-specific soil/climate. It shows the top 5 crops by advisory score (suitability, risk, and regional potential), with estimated production in kg, market price in ₹/kg, risk and disease information, and prevention measures. No profit figures are shown—advisory only. Built for final-year / academic use.

---

Features

• Region-first flow: Select state, district, and land size (bigha). No manual soil/climate input—the system uses state-specific agro-climatic defaults so recommendations vary by region (e.g. Rajasthan vs Kerala vs Himachal Pradesh).
• Advisory-only output: Top 5 crops ranked by a balanced advisory score (suitability + regional potential − risk). No net profit, ROI, or profit charts.
• Indian units: Production and sale quantity in kg; prices in ₹/kg; land in bigha (with acres shown). No quintals or tons.
• Per-crop details: Suitability %, estimated production (kg), market price (₹/kg), estimated sale quantity (kg), risk score, disease/pest risks, prevention measures, and soil-based growing tips.
• Soil nutrient view: After analysis, a soil nutrient distribution (N, P, K) chart is shown as a crop-average reference for the selected region.
• Data used for analysis: Sidebar shows total records, number of states, and total crops used by the engine. Optional refresh via data.gov.in API.
• Dark theme UI; optional lighter theme in code.

---

Problem Statement

Choosing the wrong crop for a given region and land leads to lower yield and wasted effort. This system acts as a decision-support tool: given state, district, and land size, it uses region-specific soil/climate profiles and an ML model to recommend the most suitable crops, with explainability, risk, and preventive advice—without showing direct profit to avoid misleading estimates.

---

Methodology

1. Data

   • Crop recommendation dataset: N, P, K, temperature, humidity, ph, rainfall → crop label (e.g. Kaggle – Crop Recommendation Dataset).
   • Adding more training data: Put any extra CSVs with the same columns (N, P, K, temperature, humidity, ph, rainfall, label) in data/raw/. The pipeline merges all compatible CSVs when you run python run_pipeline.py, so you can keep Crop_Recommendation.csv and add e.g. crop_extra.csv.
   • Regional data (optional): state_wise_yield.csv, market_prices.csv, cost_of_cultivation.csv, climate_vulnerability.csv in data/raw/ for state/district-aware yield, price, and risk. If absent, embedded national averages are used.

2. Region-specific inputs

   • States are mapped to agro-climatic zones (arid NW, eastern humid, southern, west coast, central, Himalayan, western dry). Each zone has distinct default N, P, K, temperature, humidity, ph, rainfall (aligned with training data). State + district offsets ensure different regions get meaningfully different inputs so recommendations vary by location and land size.

3. ML pipeline

   • Preprocessing: Label encoding, StandardScaler on training set, stratified train–test split.
   • Models: Decision Tree, Random Forest, KNN, SVM, Logistic Regression—tuned via GridSearchCV; best model by test F1-macro (e.g. SVM).
   • Prediction: predict_crop(N, P, K, temperature, humidity, ph, rainfall, land_size_bigha, state, district, scoring_mode="balanced") returns top 5 crops with suitability, production (kg), price (₹/kg), risk, disease risks, prevention, and explanations.

4. Explainability & soil health

   • Feature importance (in models/metadata.json), explanation text, and rule-based soil health messages and crop-specific suggestions.

---

Algorithms Used

Component	Role
StandardScaler	Feature scaling
LabelEncoder	Crop labels
SVM / KNN / RF / etc.	Classification (best model saved)
GridSearchCV	Hyperparameter tuning
Stratified K-Fold	Cross-validation
Region data loader	State/district yield, price, cost, vulnerability
Balanced scoring	Suitability + regional potential − risk (no profit in UI)

---

Project Structure

SMART CROP REC/
├── data/raw/              # Crop_Recommendation.csv (or sample); optional: state_wise_yield, market_prices, cost_of_cultivation, climate_vulnerability
├── models/                # model.joblib, scaler.joblib, label_encoder.joblib, metadata.json (after run_pipeline.py)
├── reports/figures/       # EDA and evaluation plots
├── src/                   # config, data_loader, zone_soil, preprocess, train, evaluate, predictor, region_data_loader, profit_engine, risk_engine, soil_health, explainer, market_price_fetcher
├── app.py                 # Streamlit UI — Smart Agriculture Advisory System
├── run_pipeline.py        # One-command ML pipeline
├── tests/                 # Crop variety tests (state, district, land size)
├── requirements.txt
├── README.md
├── REPORT.md
└── docs/ARCHITECTURE.md

---

How to Run

One-time setup

cd "SMART CROP REC"
pip install -r requirements.txt

Dataset

• Place Crop_Recommendation.csv in data/raw/ (e.g. from Kaggle), or use Crop_Recommendation_sample.csv for quick testing.
• Optional: Add state_wise_yield.csv, market_prices.csv, cost_of_cultivation.csv, climate_vulnerability.csv for better region-aware results.

Train the model

python run_pipeline.py

This loads data, runs EDA, preprocesses, trains and compares models, selects the best, and saves artifacts to models/.

Run the web app

streamlit run app.py

Or:

python -m streamlit run app.py

Then open the URL (e.g. http://localhost:8501). Select state, district, and land size (bigha) → click Proceed to Analysis → view top 5 crops, production (kg), price (₹/kg), risk, diseases, and prevention. Use Start new analysis to run again.

Run tests

python -m pytest tests/test_crop_variety.py -v

Tests verify that crop recommendations vary by state, district, and land size (not the same 5 crops for all regions).

---

Documentation

• Technical design: docs/ARCHITECTURE.md
• Academic report: REPORT.md
• Code: Comments in src/ and app.py

---

License and Dataset

Use the dataset in accordance with its source (e.g. Kaggle) and cite it in your report. This project is for academic and educational use.