A production-ready machine learning web application that recommends the optimal crop based on soil nutrients and climate conditions.
- Multi-model training — 6 algorithms trained and compared automatically
- REST API — clean Flask endpoints with full input validation
- Interactive UI — responsive web interface with animated confidence bars
- Top-5 predictions — confidence scores for the top 5 candidate crops
- Model transparency —
/api/model-infoexposes accuracy, CV scores, and feature importances - 22 crops supported across cereals, legumes, fruits and cash crops
| Layer | Technology |
|---|---|
| Backend | Python 3.10, Flask 3 |
| ML | scikit-learn 1.5 |
| Data | NumPy, Pandas |
| Frontend | Vanilla HTML/CSS/JS |
crop-recommendation/
├── app.py # Flask API & routes
├── requirements.txt
├── README.md
├── model/
│ ├── train_model.py # Training pipeline
│ ├── model.pkl # Saved best model (generated)
│ ├── scaler.pkl # StandardScaler (generated)
│ ├── encoder.pkl # LabelEncoder (generated)
│ ├── model_meta.json # Performance stats (generated)
│ └── crop_data.csv # Training data (generated)
└── templates/
└── index.html # Web UI
# 1. Clone & enter directory
git clone https://github.com/Randa-Lakab/Crop-recommendation.git
cd crop-recommendation
# 2. Create virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# 3. Install dependencies
pip install -r requirements.txt
# 4. Train the model (generates model artifacts)
python model/train_model.py
# 5. Start the server
python app.py{
"N": 80, "P": 45, "K": 40,
"temperature": 24.5,
"humidity": 85.0,
"ph": 6.2,
"rainfall": 200.0
}Returns recommended crop, confidence score, top-5 candidates, and crop metadata.
| Feature | Description | Range | Unit |
|---|---|---|---|
N |
Nitrogen | 0–140 | kg/ha |
P |
Phosphorus | 5–145 | kg/ha |
K |
Potassium | 5–205 | kg/ha |
temperature |
Avg temperature | 8–43 | °C |
humidity |
Relative humidity | 14–99 | % |
ph |
Soil pH | 3.5–9.9 | — |
rainfall |
Annual rainfall | 20–300 | mm |
| Model | Test Accuracy | CV Accuracy |
|---|---|---|
| Random Forest | ~99% | ~99% |
| Gradient Boosting | ~98% | ~98% |
| SVM | ~97% | ~97% |
| KNN | ~96% | ~96% |
| Decision Tree | ~95% | ~95% |
| Naive Bayes | ~88% | ~87% |
MIT License - see LICENSE
Randa LAKAB · LinkedIn · GitHub
Built as a Data Science project demonstrating end-to-end ML deployment with Flask.
