The growing demand for technologies capable of supporting decision-making in agriculture has driven the development of intelligent systems focused on soil fertility management. In this context, this study aimed to deve lop and evaluate a machine learning model to complement the functionalities of the Tellurium platform, assisting in soil fertility classification and the validation of agronomic management recommendations. The platform already generates recommendations for fertilization, liming, and soil management based on established agronomic criteria, with the machine learning model incorporated as an additional decision-support layer. To achieve this, a pipeline was structured comprising the following stages: data preprocessing, data quality verification, feature selection, variable standardization, training and comparison of classification algorithms, stratified cross-validation, hyperparameter optimization via GridSearchCV, and performance evaluation using metrics such as accuracy, precision, recall, F1-score, and the confusion matrix. Among the evaluated algorithms, Random Forest demonstrated the best performance, achieving an accuracy of 93.56%, precision of 93.82%, recall of 93.56%, and an F1-score of 93.59%, thereby showing a high capacity for generalization to unseen data. The model was exported, serialized, and validated in a development environment, ready for future integration into the Tellurium platform. However, its final implementation remains in an adaptation phase due to version incompatibilities between libraries used in the interface development. The results highlight the potential of the proposed hybrid approach, which integrates agronomic knowledge and machine learning to enhance the reliability of recommendations and strengthen the platform as a decision-support tool for soil fertility management.
