🌾 Crop Yield Prediction Project

This project involves a complete data science and machine learning pipeline to predict crop yield (hg/ha) using features such as average rainfall, temperature, pesticide usage, crop type, and country. Various regression models were compared, and the best-performing model was exported for deployment.

---

📊 Dataset Overview

Feature	Description
Area	Country name
Item	Type of crop (e.g. Wheat, Maize, etc.)
Year	Year of data
hg/ha_yield	Crop yield (target variable)
average_rain_fall_mm_per_year	Average rainfall (mm/year)
pesticides_tonnes	Pesticides used (in tonnes)
avg_temp	Average annual temperature (Celsius)

• Initial rows: 28,242
• Final cleaned rows: 25,932

---

🚮 Data Cleaning Highlights

• Removed Unnamed: 0 column
• Handled missing values and validated numerical types
• Dropped 2310 duplicate records
• Explored distribution using Seaborn plots

---

📊 Exploratory Data Analysis (EDA)

• Yield distribution across countries and crop types visualized using bar plots

• Top 3 highest yield crops:

  • Potatoes
  • Sweet potatoes
  • Cassava

• Yield analysis done per Item and Area

---

📅 Feature Engineering

• Selected key features:

  • Year, average_rain_fall_mm_per_year, pesticides_tonnes, avg_temp, Area, Item

• Target: hg/ha_yield

• Applied preprocessing using ColumnTransformer:

  • OneHotEncoding for categorical (Area, Item)
  • StandardScaler for numeric features

---

🔧 Models Used & Performance

Model	Mean Absolute Error (MAE)	R2 Score
Linear Regression	29,920.77	0.75
Lasso	29,907.60	0.75
Ridge	29,875.65	0.75
Decision Tree	4,206.54	0.98
KNN	4,869.09	0.98

✅ Decision Tree Regressor performed the best and was chosen for deployment.

---

🔮 Predicting Crop Yield (Example)

from predict import predict_yield

predict_yield([2023, 1200, 50, 30, 'Pakistan', 'Wheat'])
# Output: 27957.0 hg/ha

---

🚀 Deployment

The following assets are exported using pickle for easy deployment in web apps (e.g., FastAPI):

• website/crop_yield_model.pkl → Trained Decision Tree model
• website/crop_yield_preprocessor.pkl → Preprocessing pipeline

---

🧰 Tech Stack

• Pandas, NumPy – Data manipulation and analysis
• Seaborn, Matplotlib – Visualizations
• Scikit-learn – Preprocessing, regression models, metrics
• Pickle – Saving models for deployment

---

📂 Project Structure

.
├── yield_df.csv                  # Raw dataset
├── crop_yield.ipynb             # Full notebook
├── predict.py                   # yield prediction function
├── website/
│   ├── crop_yield_model.pkl     # Exported Decision Tree model
│   └── crop_yield_preprocessor.pkl  # Exported ColumnTransformer

---

📉 Future Improvements

• Add rainfall forecasts and satellite vegetation index
• [✅] Deploy via FastAPI
• Add dynamic crop recommendation based on yield optimization

---

🙌 Acknowledgments

Developed by Haris Ahmed as part of AI/ML project portfolio. Data cleaned, analyzed, modeled, and deployed end-to-end.