A comprehensive analysis of feature selection and XAI for machine learning classifiers to recognize guava disease

Abstract

Recognizing and classifying diseases in guava is crucial for managing farms to keep crops healthy and increase harvest quality. Cultivators face the most severe challenges when it comes to recognizing and diagnosing guava fruit and leaf illnesses, a task that is nearly impossible to perform manually. This research focuses on developing a robust disease identification model using image data collected locally from guava trees. After data collection, various image processing techniques, including scaling and contrast enhancement, are utilized to make the data more suitable for use. K-means clustering is employed to quickly divide the images into groups, followed by the extraction of important characteristics. Two separate feature ranking approaches, analysis of variance (ANOVA) and least absolute shrinkage selection operator (LASSO), are used to select the best characteristics, identifying the 10 most important attributes. The adaptive boosting (AdaBoost) classifier achieves the highest accuracy among six classifiers for the top seven characteristics indicated by LASSO among the specified features. To enhance the model’s interpretability, two explanation methods, local interpretable model-agnostic explanations (LIME) and shapley additive explanations (SHAP), are employed to illustrate how the classifier reaches its conclusions. This approach not only simplifies disease identification but also clarifies the reasoning behind predictions, opening the door to real-world applications in detecting and preventing dangerous diseases.