🌾 Crop Yield Prediction using Machine Learning

📌 Project Overview

Agriculture is a key sector in when it comes to our country, and predicting crop production can help optimize resource allocation, yield forecasting, and policy-making. In this project, I used historical crop production data to build predictive models that estimate crop yield based on features like State, District, Crop, Season, Area, and Year.

Accurate crop prediction can help in agriculture planning, resource optimization, and policy-making.

---

📊 Dataset

• Source: Agriculture Crops Production in India - Kaggle
• Rows: 345,407
• Columns: 10
• Target Variable: Yield

Key Features:

• State - Name of the state
• District - Name of the district
• Season - Season of cultivation
• Crop - Type of crop grown
• Year- Year of cultivation
• Area - Cultivation area (in hectares)
• Production - Total crop production
• Yield - Target Variable

---

✅ Project Workflow

1. Dataset Overview
   • Read Dataset
   • Display first 5 rows
   • Shape & Structure
2. Exploratory Data Analysis (EDA)
   • Null value analysis
   • Correlation heatmap of numerical features
   • ANOVA test for categorical columns
   • Production distribution analysis
   • Top crops by production
   • District-wise and season-wise crop production
3. Data Preprocessing
   • Imputation using SimpleImputer
   • Cardinality Calculation for Categorical Variables
   • One-hot encoding for low-cardinality variables
   • Frequency encoding for high-cardinality variables
4. Model Development
   • Model 1: Linear Regression (Baseline)
   • Model 2: Ridge Regression (with RandomizedSearchCV tuning)
   • Model 3: Lasso Regression (with RandomizedSearchCV tuning)
   • Model 4: XGBoost Regressor
     • Manual tuning (7 different parameter sets)
     • Hyperparameter tuning using RandomizedSearchCV
5. Model Evaluation
   • Metrics: R², RMSE, MSE, MAE
   • Actual vs Predicted scatter plots for all models
6. Feature Importance
   • Extracted feature importance from XGBoost

---

🤖 Models Implemented & Performance

Model	R² Score	Observations
Linear Regression	~0.79	Baseline
Ridge Regression	~0.79	Similar to LR
Lasso Regression	~0.79	Similar to Ridge
XGBoost Regressor	0.945	Best performing model

---

📈 Key Visualizations

• Correlation heatmap of numerical features
• Seasonal and state-wise crop production trends
• Top crops by production
• Actual vs Predicted plots for all models
• XGBoost feature importance chart

---

✅ Conclusion

• XGBoost Regressor was the best-performing model with an R² score of 0.945 and the lowest RMSE.
• Regularization (Lasso/Ridge) did not significantly improve results compared to Linear Regression.
• Top 5 crops by production: Coconut, Sugarcane, Rice, Wheat, Potato.
• Highest production seen in states: Kerela, Tamil Nadu, Karnataka, Andhara Pradesh, West Bemgal.
• Seasonal trend shows crops cultivated whole year leads in overall production while Kharif and Rabi come after it.

---

🌍 Real-World Applications

• Assisting farmers and policymakers in making informed decisions on crop planning.
• Optimizing resource allocation based on expected production.

---

🚀 Future Enhancements

• Incorporate weather and soil data for better accuracy.
• Deploy the model using Flask, FastAPI, or Streamlit for real-time prediction.
• Build a dashboard for visualization and decision support.

---