Green gram yield prediction using linear regression

Abstract

Predicting crop yields before harvest is essential for farmers to make informed decisions about post-harvest management and for guiding agricultural enterprise selection, ultimately promoting food and nutrition security. Crop yields are influenced by diverse factors, including ecological zone characteristics and farm management practices, which vary significantly across seasons and farmers. Despite the growing adoption of yield prediction approaches such as satellite imaging and plant physiology analysis, limited studies exist for neglected crops like green gram compared to popular crops like rice and maize. This study proposes a green gram yield prediction model using stepwise linear regression. Key variables include ecological zone characteristics, farm management practices, and historical yield data. The model was developed using a dataset of 107 gardens and nine features collected from the National Semi-Arid Research Institute (NaSARRI), Serere, Uganda. Predictor variables included soil type, pH, fertility, rainfall distribution, weeding practices, pest and disease management, fertilizer use, plant spacing, and cropping systems. The model’s performance was evaluated using mean absolute percentage error (MAPE), achieving 16.8%, and a precision of 96.4%, demonstrating its accuracy in predicting green gram yields. This research contributes to addressing the knowledge gap in yield prediction for neglected crops, offering a practical tool for farmers and agricultural planners in semi-arid regions to optimize productivity and resource allocation.