Explainable Regression Models for Crop Productivity Prediction

The precise forecasting of crop yield is of immense importance in contemporary agriculture, as it helps farmers and policymakers make wise decisions about resource allocation, irrigation schedules, and the use of fertilizers. In this research, a machine learning-based approach for crop yield forecasting based on formal agricultural data sets is proposed. The method makes use of historical and intelligent farming data, combining factors like soil type, climatic conditions, irrigation techniques, and farm practices to create a predictive model that calculates an estimate of yield in kilograms per hectare. The system makes use of a comprehensive pipeline starting with data auditing and preprocessing, including missing value handling, categorical variable encoding, and feature engineering, like extracting sowing-to-harvest time frames. Random Forest Regressor is the baseline model with excellent performance metrics in all R2, MAE, and RMSE. In addition, model explainability is incorporated in the framework using SHAP values and providing both global and local feature contribution insights. A final enrichment step includes external agricultural data and enhances the model accuracy and proves the benefit of data fusion for precision agriculture. To aid usability and interpretability, the project integrates a what-if analysis module that predicts the effect of varying agronomic parameters, like rainfall, fertilizer amount, and irrigation scheme, on forecasted yield.