Enhancing Agricultural Productivity with Machine Learning-Based Soil Fertility Prediction

Context: Soil fertility plays a vital role in the agricultural field, as it provides a descriptive analysis of the nutrient content present in the soil and helps crops grow well in that soil. According to various sources, the majority of farmers are unaware of their soil fertility type. As a result, they face losses when harvesting crops because the type of crop may not be suitable for their preferred soil. Objective: The primary objective of this paper is to predict different levels of soil fertility by analyzing the nutrient content present in the soil using machine learning. Materials and methods: In this study, several machine learning classification algorithms—namely, Gradient Boosting, Extra Trees, Bagging Classifier, Random Forest, Decision Tree, AdaBoost, and SVM—were individually trained and evaluated on the same soil dataset to compare their predictive performance. Each classifier was employed to learn patterns from the selected soil features and predict the fertility class of each soil sample. The comparative analysis revealed that the Gradient Boosting classifier produced the most accurate predictions (Accuracy – 0.9806). Apart from this, other algorithms, such as Random Forest, Decision Tree, and Extra Trees, provided nearly equal results. However, even after hyperparameter tuning, the Gradient Boosting algorithm still provides the best result among all the algorithms.