Hybrid Paddy disease classification using optimized statistical feature based transformation technique with explainable AI

Abstract Agricultural productivity is at risk from plant diseases all across the world, but it is particularly harmful in countries that are developing where farming is crucial to the economy. Avoiding crop failure due to microbes, viruses, fungi, or environmental stress requires the prompt detection of paddy leaf diseases. Traditional disease classification relies on expert knowledge, which farmers may not always have access to. This research proposes a machine learning strategy based on Explainable AI that can assist farmers quickly diagnose and manage rice leaf diseases. To enhance predictive performance by transforming features, an improved Owl Search Optimization (IOSO) algorithm is used. This method not only yields consistent outcomes, but it also provides insights into its decisions, which helps the user better understand and trust the data. The Yeo-Johnson technique is used to alter the statistical information acquired from each image. The next step is to use a Cat Boost classifier to sort the features. Using SHAP analysis and the Cat Boost classifier, this study delves deeper into the model’s classification process and the effects of each feature on the model’s operation. The ideal transformation method substantially improved performance, leading to an accuracy of up to 98.76%, even though the original model showed a balanced accuracy of less than 75%. This method proved to be the most effective in correctly identifying paddy leaf illnesses, surpassing other models such as SVM, Random Forest, and XGBoost.