Machine Learning Techniques for Cereal Crops Yield Prediction: A Comprehensive Review

Abstract

Cereals are sensitive to small changes in complex combinations of biotic and abiotic factors. Such a complexity can be deciphered using techniques such as Machine learning (ML). Using the PRISMA approach, this paper explores the features and ML techniques in cereal yield prediction based on 115 articles from 2007 to 2023 in six databases. Results showed that most data in the articles were from secondary sources and only 28.68% used experiments or primary data. China (31) and the United States (18) contributed most. Wheat (48%), maize (33%), and rice (17%) represented the most studied cereals. Climate, remote sensing data, and soil parameters were the most used predictors. The most frequently used ML techniques for cereal prediction were support vector machine (SVM) (51%), multi-layer perceptron (MLP) (41%), linear regression (34%), random forest (RF) (24%), and XGBoost (20%). However, RF, MLP, and SVM models were the best-performing techniques to predict grain yield based on reported R-square and mean absolute error (MAE). The models in the studied articles generally performed well from test data, with an R-square between 0.7 and 1. The study further reveals that the data's availability and quality are the main obstacles to using ML models for crop prediction.

Keywords

• cereal
• deep learning
• machine learning
• prisma
• yield prediction