IMPROVING PLANT DISEASE CLASSIFICATION WITH DEEP-LEARNING-BASED PREDICTION MODEL USING EXPLAINABLE ARTIFICIAL INTELLIGENCE

Abstract

Plant diseases pose significant threats to agricultural productivity, affecting both crop yields and quality. These diseases have a profound economic impact o n local and global scales, causing substantial losses in food security and trade. Deep learning algorithms have emerged as effective solutions for plant disease detection and classification, offering advanced capabilities for accurate and automated analysis. However, the inherent "black-box" nature of these models raises concerns about the interpretability and trustworthiness of their predictions, highlighting the need for integrating explainable artificial intelligence (XAI) to enhance user confidence in model decisions. This study addresses these challenges by evaluating deep learning techniques for plant disease classification and detection using the Plant Leaf Disease dataset. For classification, algorithms such as Convolutional Neural Networks (CNN), MobileNetV2, ResNet50, EfficientNetB0, Xception, and NasNetMobile are compared, alongside an ensemble model combining Xception and NasNetMobile, which demonstrated superior performance. Performance is evaluated using metrics such as recall, precision, and F1 score, which provide a comprehensive understanding of model effectiveness. The ensemble approach of Xception and NasNetMobile achieved the highest accuracy in classification, while the YOLO models proved robust in detecting plant diseases. This research underscores the importance of incorporating XAI with deep learning to foster trust in AI systems and improve agricultural disease management practices.