Insights of Machine Learning Approach for Soil Fertility Assessment and Management Strategy.

Soil is a vital resource that serves as a foundation for sustainable food production by providing the required physical, chemical, and biological characteristics required for healthy crop. But, the quality of soil is diminishing due to repeated cultivation without proper knowledge. Hence, assessing soil fertility is crucial for decision making and management. To effectively manage the soil, farmers need to know its composition, nutrient level, and other characteristics. One important instrument for this process is soil testing. Laboratory analysis of soil parameters despite being labor-consuming and time-intensive process, it also leaves chemical residues that pollute the ecosystem. Using nonchemical techniques for analyzing soil nutrients through machine learning (ML) offers efficiency, and scalability and helps in real-time decision-making. Integrating ML in agriculture can better solve complicated issues and give producers essential advice. ML algorithms have been used to generate forecasting models based on available soil data and environmental variables. This review highlights the models used in soil fertility assessment and their performance and accuracy rates across different applications. A summary of the major findings from the studies discussed shows that neural networks such as artificial neural networks (ANN) performed well in phosphorus prediction with 91.2% accuracy, convolutional neural networks (CNN) achieved 98.4% in nutrient prediction. Random forests (RF) achieved an accuracy of up to 99% in soil organic carbon (SOC) prediction. Ensemble Learning like XGBoost in pH prediction and crop recommendation and AdaBoost in potassium prediction shows the highest accuracy reaching up to 99.3%. This review along with similar findings further explores the importance of ML in soil fertility assessment. [ABSTRACT FROM AUTHOR]