Crop Yield Prediction

A robust machine learning model that predicts crop yields using key agricultural features like crop type, state, season, area, rainfall, pesticide, and fertilizer usage. This model is designed to assist farmers, agricultural planners, and researchers in making data-driven decisions to maximize productivity and sustainability.

---

Features

• CatBoost Regrassor for Prediction
• Explainable AI with SHAP for transparency
• Hyper Parameter Tuning using Optuna
• Log-transformation for stable and normalized target distribution
• Insight into which features impact yield the most

---

Model

The notebook (model.ipynb) builds a regression model based on Categorical Boosting where the target variable is Yield (computed as Production / Area). It uses structured agricultural data with features such as:

• Crop
• Crop_Year
• Season
• State
• Area
• Production
• Annual_Rainfall
• Fertilizer
• Pesticide

---

Training & Evaluation

• To improve prediction stability and normalize skewed yield distribution, we apply a log transformation to the target variable:

    ```python
    df = df[df['Yield'] > 0]
    df['log_yield'] = df['Yield'].apply(lambda x: np.log(x))
  
    # similarly with Fertilizer and Pesticides per Area
    ```
  

• Data split (4:1) into training and testing sets

• Evaluation Metrics

  Metric	Value
  Mean Absolute Error (MAE)	16.7743
  Median Absolute Error	0.1770
  Median Absolute Percentage Error	0.3099
  R² Score	0.9553
  Adjusted R² Score	0.9552

---

Plots

Prediction Error Vs True Yield

Shap Plot

Optuna Hyperparameter Tuning

---

Dataset: https://www.kaggle.com/datasets/akshatgupta7/crop-yield-in-indian-states-dataset