This paper proposed an ensemble learning algorithm to enhance classification accuracy in agricultural dataset based on S-approximation. In the field of agriculture, the use of these algorithms can provide valuable insights into crop growth patterns, disease outbreaks and other important factors. Four ensemble learning methods, including Random Forest, boosting, bagging and majority voting, were employed and were used with a real agricultural dataset. The proposed method identifies the dependency degree of attributes and selects reducts that are most relevant for classification tasks. These reducts are subsequently used to train an ensemble of base classifiers. This hybrid approach combines the strengths of S-approximation with ensemble learning techniques to provide a robust classification framework. To evaluate the proposed algorithm, experiments were conducted on a real-world crop yield dataset, categorizing crop yields into “Low”, “Medium”, and “High” production levels. Production and Fertilizer, were identified as critical predictors. The dataset was pre-processed to address missing values, normalize attributes and to address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was employed, ensuring equitable representation of all classes. The result demonstrates that Random Forest achieved the highest accuracy (99.08\%) with balanced precision, recall and F1-score, confirming its robustness and reliability. The Voting Classifier, integrating the strengths of individual models, also delivered high accuracy (99.04\%), showcasing the efficacy of the ensemble approach. This work highlights the potential of integrating S-approximation with ensemble learning to improve classification tasks in agricultural datasets, providing actionable insights for optimizing crop yields and supporting data-driven agricultural decision-making.