Journal of Agronomy and Field Innovations

Background: Correct forecasting of agricultural output is crucial for food security programs, resource management, and climate-smart farming practices. The traditional methods used in statistical and process-oriented approaches for predicting crop yields fail to reveal the dynamics of non-linear links that exist between different climatic and edaphic factors responsible for the crop yield.
Objective: The purpose of the current study is to present a new prediction framework based on the SVM model which treats climate factors in combination with such soil indicators as moisture, pH, and content of nitrogen, phosphorus, potassium, organic carbon.
Methods: The agricultural dataset that contained climate, soil, and historical yield data was processed by means of missing value imputation, outlier removal, scaling of features, and selection of features using recursive algorithms. SVM regression model with a radial basis function (RBF) was trained on the basis of grid-search hyperparameter tuning, 10-fold cross validation, and was compared to Decision Tree, Random Forest and K-Nearest Neighbor (KNN) approaches.
Results: The SVM model indicated a R² value of 0.91, RMSE value of 0.38 tons per hectare, and MAE value of 0.29 tons per hectare which is better than baseline models in terms of accuracy and generalization for prediction. It was also found that soil nitrogen and seasonal rainfall played important role as predictors in the SVM model.
Conclusion: SVM based modeling provides a strong and computationally efficient tool for precision agriuculture in real-life if used in combination of climatic variables with soil data.