AI-Based Soil & Sensor Data Driven Crop Yield & Health Prediction

(Mini Project — Precision Agriculture / AIML)

Overview

This mini project predicts crop yield (kg/hectare) and crop health status (Healthy / Stressed / Poor) from soil and environmental IoT sensor readings (N, P, K, pH, soil moisture, temperature, humidity, rainfall) using Random Forest machine learning models.

Folder Structure

mini_project/
├── data/
│   └── sensor_dataset.csv        # Generated synthetic sensor dataset (1200 samples)
├── models/                       # Saved trained models + scalers (.pkl)
├── outputs/                      # Evaluation plots + metrics summary
├── src/
│   ├── generate_dataset.py       # Creates the synthetic sensor dataset
│   ├── train_model.py            # Preprocessing, training, evaluation, plots
│   └── predict_sample.py         # Demo: predict yield & health for new readings
└── README.md

How to Run

pip install scikit-learn pandas numpy matplotlib seaborn joblib

python src/generate_dataset.py    # Step 1: create dataset
python src/train_model.py         # Step 2: train + evaluate models
python src/predict_sample.py      # Step 3: run inference on sample sensor readings

Models

• Yield Prediction: RandomForestRegressor
• Health Classification: RandomForestClassifier (Healthy / Stressed / Poor)

Notes

The dataset is synthetically generated with agronomically-informed relationships (e.g., yield rises with balanced NPK, pH near neutral, moderate moisture/temperature) plus realistic noise, since live multispectral/IoT field data was not available for this mini-project scope. The same pipeline works unmodified on real sensor data — only generate_dataset.py would be swapped for a real data loader.