Yield prediction for crops by gradient-based algorithms

• Published in: PloS one Vol. 19; no. 8; p. e0291928
• Main Authors: Mahesh, Pavithra, Soundrapandiyan, Rajkumar
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 26.08.2024; Public Library of Science (PLoS)
• Subjects: Agricultural commodities; Agricultural economics; Agricultural equipment; Agricultural industry; Agricultural production; Agriculture; Agriculture - methods; Algorithms; Analysis; Artificial intelligence; Crop yield; Crop yield forecasting; Crop yields; Crops; Crops, Agricultural - growth & development; Decision making; Environmental factors; Forecasts and trends; International trade; Learning algorithms; Machine Learning; Management; Oryza - growth & development; Pesticides; Rainfall; Seasonal variations; Soil fertility; Soil temperature; Soil water; Technology application; Temperature; Water availability; Water quality; India
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0291928

A timely and consistent assessment of crop yield will assist the farmers in improving their income, minimizing losses, and deriving strategic plans in agricultural commodities to adopt import-export policies. Crop yield predictions are one of the various challenges faced in the agriculture sector and play a significant role in planning and decision-making. Machine learning algorithms provided enough belief and proved their ability to predict crop yield. The selection of the most suitable crop is influenced by various environmental factors such as temperature, soil fertility, water availability, quality, and seasonal variations, as well as economic considerations such as stock availability, preservation capabilities, market demand, purchasing power, and crop prices. The paper outlines a framework used to evaluate the performance of various machine-learning algorithms for forecasting crop yields. The models were based on a range of prime parameters including pesticides, rainfall and average temperature. The Results of three machine learning algorithms, Categorical Boosting (CatBoost), Light Gradient-Boosting Machine (LightGBM), and eXtreme Gradient Boosting (XGBoost) are compared and found more accurate than other algorithms in predicting crop yields. The RMSE and R 2 values were calculated to compare the predicted and observed rice yields, resulting in the following values: CatBoost with 800 (0.24), LightGBM with 737 (0.33), and XGBoost with 744 (0.31). Among these three machine learning algorithms, CatBoost demonstrated the highest precision in predicting yields, achieving an accuracy rate of 99.123%.