A Review on Machine Learning Model for Predicting Maize Crop Yield in Semi-Arid Regions

Abstract

Crop yield prediction is an important task for achieving global food security and optimizing agricultural resource use in the face of climate change and land degradation. Machine learning (ML) models are able to capture the non-linear dynamics in crop systems, but their application is limited by crop and region specific dependencies. This review critically assesses the strengths, limitations and generalizability of hybrid ML models for yield forecasting of maize in semi arid regions. Six major academic databases were searched for 64 peer reviewed studies that were relevant, methodologically rigorous and used hybrid ML frameworks. The studies were grouped by model type, data sources, application domains, and generalization techniques. Results show that the integration of ML, DL, and metaheuristic algorithms improves the predictive accuracy, especially in data-scarce conditions. Multimodal data fusion, contextual feature selection, and algorithmic diversity are key performance drivers. Moreover, transfer learning and domain adaptation techniques greatly enhance cross crop model portability. Thus, hybrid ML models become important tools for predictive agriculture. Nevertheless, future efforts should focus on data standardization, explainable model architectures, and infrastructure accessibility to unleash their full potential. To support decision making across a range of agroecological systems, particularly in fragile environments, it is essential to build modular, scalable, crop agnostic models.