Yield prediction for crops by gradient-based algorithms.

Area of Science:

• Agricultural Science
• Data Science
• Machine Learning

Background:

• Accurate crop yield assessment is crucial for farmer income, loss minimization, and strategic agricultural planning.
• Crop yield prediction is a significant challenge in agriculture, impacting decision-making and policy.
• Environmental and economic factors influence crop selection and yield.

Purpose of the Study:

• To evaluate and compare the performance of various machine learning algorithms for crop yield forecasting.
• To identify the most accurate machine learning model for predicting crop yields based on key parameters.

Main Methods:

• Developed forecasting models using machine learning algorithms: Categorical Boosting (CatBoost), Light Gradient-Boosting Machine (LightGBM), and eXtreme Gradient Boosting (XGBoost).
• Utilized parameters including pesticides, rainfall, and average temperature for model training.
• Calculated Root Mean Square Error (RMSE) and R-squared (R2) values to assess prediction accuracy against observed rice yields.

Main Results:

• CatBoost achieved the highest precision with an accuracy rate of 99.123%.
• RMSE and R2 values for CatBoost were 800 (0.24), LightGBM were 737 (0.33), and XGBoost were 744 (0.31).
• Compared to other algorithms, CatBoost, LightGBM, and XGBoost demonstrated superior accuracy in crop yield prediction.

Conclusions:

• Machine learning algorithms, particularly CatBoost, show significant promise for accurate crop yield prediction.
• The study framework provides a reliable method for evaluating ML model performance in agriculture.
• Accurate yield forecasts can support farmers and policymakers in agricultural commodity management.