EDXRF and Machine Learning for Predicting Soil Fertility Attributes

Soil fertility evaluation is fundamental for sustainable agricultural practices, often relying on conventional laboratory methods. These methods, while accurate, are labor-intensive, time-consuming, and require chemical reagents. Spectroscopic sensors, such as energy-dispersive X-ray fluorescence (EDXRF), offer a rapid and non-destructive alternative but require calibration of machine learning models for accurate prediction of fertility attributes. In this context, this study compares the performance of four machine learning algorithms—multiple linear regression (MLR), partial least square regression (PLS), support vector machine regression (SVM), and random forest regression (RF)—in predicting soil pH, organic carbon (SOC), sum of exchangeable bases (BS), and cation exchange capacity (CEC) using EDXRF data from two soil datasets. Results indicate that PLS models outperformed others (the hierarchy of accuracy was PLS > MLR > SVM > RF). Overall, we emphasize the benefits of integrating PLS with EDXRF, capable of mitigating the use of traditional soil analysis.