A Novel Image-Based Machine Learning Approach for Feature-Driven Plant Disease Detection

Plant disease detection significantly impacts precision agriculture by enabling early and accurate disease identification, as well as timely and targeted treatments, resulting in substantial reductions in crop losses and optimized resource utilization. Concerning plant disease detection, deep learning methods frequently encounter limitations stemming from their high computational costs and exacting technical requirements. Therefore, this study utilizes machine learning techniques over deep learning to reduce costs and effectively manage the technical complexities involved. This research introduces a novel Stacking Ensemble Classifier model for enhancing the accuracy of Apple leaf disease diagnosis. By integrating four base classifiers—Random Forest, Gradient Boosting, XGBoost, and LGBM—the Stacking Ensemble Classifier achieves notable improvements in prediction accuracy through robust image preprocessing and extensive hyperparameter tuning. The study leverages a dataset of over 5700 Apple leaf images sourced from PlantVillage. Results demonstrate that the proposed model achieved superior performance, surpassing individual existing machine learning classifiers with an outstanding accuracy of 98.44%. This approach maintains high detection rates while reducing computational overhead, outperforming other existing classifiers in accuracy as well as significant performance parameters comprising precision, recall, and F1-score.