Optimizing Crop Yield Prediction Using C4.5 Classification Algorithm for Sustainable Agriculture

Accurate crop yield prediction is a critical component of sustainable agricultural planning, particularly in regions affected by climatic variability and limited resources. This study evaluates the effectiveness of the C4.5 decision tree classification algorithm for crop yield prediction using agro-climatic and soil-related variables. Empirical data obtained from farms in Mubi North, Adamawa State, Nigeria, comprising soil nutrients (N, P, K), pH, organic matter, rainfall, temperature, humidity, and management practices, were used to classify crop yield into low, medium, and high categories. The performance of the optimized C4.5 model was compared with ensemble learning approaches, namely Random Forest and Gradient Boosting classifiers. Model enhancement techniques, including pruning, cost-sensitive learning, and probability calibration, were applied to improve generalization and reliability. Experimental results show that the pruned C4.5 model achieved superior performance, recording 1.00 accuracy, macro-F1 score, and Cohen’s Kappa on the test dataset, outperforming the ensemble models under the given experimental conditions. Feature importance and SHAP-based explainability analysis identified rainfall and soil nutrient levels as the most influential predictors of yield variability. The findings demonstrate that interpretable decision tree models remain highly effective for crop yield prediction in data-constrained agricultural environments and provide actionable insights for sustainable farm management.