Comparing Machine Learning Algorithms for Rice Yield Prediction in Adamawa and Cross Rivers States of Nigeria

Abstract

Rice production is critical for global food security, and accurate yield prediction empowers informed decision-making. This paper investigates machine learning (ML) techniques for rice yield prediction in Adamawa and Cross River states, with distinct agroclimatic conditions. Traditional yield prediction methods that are commonly used often have limitations such as less insights into the available data and reduced accuracy. Hence, this research explores the potential of machine learning for improved prediction accuracy. We leverage climatic data and historical rice yields to train and evaluate Decision Trees, Random Forest, Support Vector Regressor, Polynomial Regressor, Multiple Linear Regression and Long Short-Term Memory (LSTM) models. Performance is compared using Mean Squared Error, Root Mean Squared Error, Coefficient of Determination, Mean Absolute Error, and Mean Absolute Percentage Error. Feature selection identifies All-sky Photosynthetically Active Radiation (PAR) as the most influential factor. Linear Regression emerges as the superior model, achieving an R² of 0.90 (Adamawa) and 0.91 (Cross River), demonstrating robust generalizability across regions. This research contributes to the development of ML-powered Agro-information systems for two Nigerian regions, enhancing agricultural practices and food security.