Predicting Rice Crop Yields in Prayagraj: A Comparison of Regression Techniques and Neural Networks

Abstract

This study investigates rice crop yield prediction in Prayagraj District, Uttar Pradesh, using twenty-nine years of historical weather and crop yield data from 1991 to 2019. The analysis divided the dataset into a calibration segment comprising 90% of the data over 26 years, with the remaining 10% used for validation. In our approach, 75.9% of the data was used to train an artificial neural network (ANN) model, and 24.1% was employed for testing and validation to ensure 100% model efficiency evaluation. Techniques such as stepwise linear regression and neural networks were applied to predict rice yields. The performance of these models was measured primarily by the Normalized Root Mean Squared Error (nRMSE), with the regression-based model achieving superior performance, indicated by the lowest nRMSE values. The study also noted the critical role of Bright Sunshine Hours, which demonstrated a significant predictive power with an nRMSE of 0.00025 and a coefficient of determination of 0.94.