Predictive crop yield analysis based on weather and soil data for agriculture optimization

Accurate and timely crop yield predictions are essential for addressing global hunger, optimizing resource use, and managing the increased farming challenges posed by extreme weather variations. This study explores four key machine-learning techniques: Linear Regression, Random Forest Regression (RFR), Gradient Boosting, and Support Vector Regressor (SVR). Conventional analytical methods often fail to effectively unravel the complex relationships between crops and nature’s fluctuations. Leveraging a substantial dataset of 28,242 farm records collected from 1990 to 2020 in regions such as South Asia and Sub-Saharan Africa, the research examines crops including maize, rice, potatoes, and wheat. It focuses on the impact of rainfall patterns, temperature variations, and pesticide application on crop yields across various farming conditions. Feature importance analysis identifies the primary factors influencing yields, while residual checks assess prediction accuracy and identify any issues. Achieving a mean squared error (MSE) of 120.9 million and a robust R² of 0.9833—indicating an excellent fit—findings reveal RFR as particularly effective for yield predictions. Temperature and pesticide usage stand out as major determinants. Visualization tools like heatmaps and residual charts illuminate trends, demonstrating that RFR can translate reliable data analysis into practical farming decisions, paving the way for more sustainable agricultural practices.