European Journal of Sustainable Agroecosystems

Background: Fertilizing wastefully and indiscriminately is a constant obstacle on the road toward sustainable farming because it creates both nutrient shortages and environmental damage by means of excessive fertilizer applications. There was a steady increase in five years of global mineral fertilizer consumption which went up from 142 million tonnes in 2002 to around 190 million tonnes in 2023, with nitrogen accounting for more than half of all consumption; that’s why it became necessary to implement site-specific approaches for managing fertilizers. Analyzing and predicting the future of crops is considered to be an innovative way to achieve a balance between production efficiency and environmental safety with the help of diverse predictive analytics technologies including machine learning (ML), deep learning (DL), IoT, remote sensing, etc.

Objective: This paper is a review of academic literature published in the span of 2015-2025 on applying predictive analytics methods in fertilizers recommendation aiming at recognizing trends in methodologies, assessing model’s efficiency and identifying research gaps.

Sources of literature: An organized search was performed using Scopus, Web of Science, PubMed, ScienceDirect, SpringerLink, MDPI, IEEE Xplore, as well as Food and Agriculture Organization (FAO) and International Fertilizer Association. Some major studies published before 2015 were also used in this review.

Major findings: 58 full-text studies were eligible and finally 39 studies were considered useful for thematic synthesis. In terms of efficiency of tree-based ensemble algorithms, both random forest and XGBoost, as well as artificial neural network methods achieved more than a 90% accuracy rate for predicting fertilizers based on crops and locations. The use of decision support systems including Nutrient Expert and the Fertilizer Recommendation Support Tool showed positive outcomes for the field level. Moreover, the application of IoT farming helped farmers get better results.

Research: There are many limitations regarding the alleged development of useful insights in relation to predictive analytics. Thus, the weak adaptability of a fixed model to real-time scenarios, low generalizability of results in terms of evidence from different regions of the world, unclear reasoning, absence of data in relation to small-scale farmers along with benchmark datasets for cross-comparisons among studies on the subject limit the research process. At the same time, predictive analytics can be effectively applied as a means for improving fertilizer recommendations and indeed achieving the goals of food security, profitability of agricultural endeavors, and environmental protection, providing that future investigations will focus on the development of models that can generate information that is comprehensible, regionally adjusted, and verified.