Machine Learning-Based Smart Agricultural Practices To Assess Soil Fertility And Nutrient Dynamics

Abstract

This research utilises machine learning (ML) to improve agricultural precision by forecasting soil characteristics levels using environmental data. A dataset that comprised additional information on temperature, humidity, soil moisture, nitrogen (N), phosphorous (P), and potassium (K) values was used to create models that proposed recommendations for adequate fertilizer application. The performance of models was assessed utilizing R-squared, Adjusted R-squared, mean absolute error (MAE), and mean squared error (MSE). The random forest (RF) model was more accurate than others, showing the lowest MSE for P (mg/kg) and competitive MAE for other characteristics. Gradient boosting models had higher errors and negative R-squared values, suggesting they didn’t fit the data, even though the results were close in performance. Linear regression proved to be reliable with the lowest MAE for N (mg/kg) and K (mg/kg) and the most significant R-squared values for P (mg/kg), showing its persuasiveness in accurately forecasting these characteristic levels despite its simplicity. The research leverages machine learning to precisely predict soil nutrients for smarter farming, with the random forest model providing superior accuracy over other techniques. These advancements highlight the importance of continuous innovation in environmental monitoring for sustainable agriculture.