Crop Yield Prediction Using Different Techniques of Machine Learning for Prayagraj Region

Abstract

Reliable crop yield prediction is becoming increasingly critical as climate patterns shift, and successful agricultural planning depends heavily on our ability to accurately forecast production. This study introduces a data-driven framework designed to untangle the complex relationship between local weather patterns and crop performance in the Prayagraj region. We focused on five key crops—Maize, Wheat, Rice, Mustard & Rapeseed, and Potato—to determine how well modern computational tools can reduce the uncertainty found in traditional assessments. We compared several methodologies to address both linear and non-linear data dependencies, ranging from regression techniques (LASSO, Elastic Net, Ridge, Stepwise MLR) to ensemble and neural network models (Random Forest, ANN). Quantitative evaluation revealed that models trained on weighted weather data generally exhibited superior stability. specifically, Artificial Neural Networks (ANN) achieved the highest predictive accuracy for Potato (nRMSE = 0.13) and Wheat (nRMSE = 0.19). For Maize, regularized regression models (Elastic Net, LASSO, Ridge) proved most effective (nRMSE = 0.14), while Random Forest (RF) demonstrated robust generalization for Mustard & Rapeseed (nRMSE = 0.18) and Rice (nRMSE = 0.20).