EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI) APPLICATION ON CLASSIFICATION OF DISEASES AND SYMPTOMS OF PEST INFESTATION ON PLANTS (CASE STUDY ON RICE PLANT LEAVES)

• Main Author: Manuel Kurniawan, William
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84978
• Institution: Institut Teknologi Bandung

Detection of disease and pest symptoms in rice plants is essential to prevent potential losses caused by plant damage. While machine learning is a common approach for detecting these symptoms, it has the drawback of functioning as a "black box," making it difficult for users to fully comprehend the underlying processes. This lack of transparency diminishes the accountability of the detection results. To address this issue, explainable artificial intelligence (XAI) offers a solution by providing insights into the detection process. This Final Project aims to develop a classification model for rice plant diseases and pest symptoms, implement an XAI framework to explain the classification results, and deploy the implementation within a mobile application. The classification model is constructed using the transfer learning method to enhance detection performance. XAI methods, including LIME, SHAP, and Grad-CAM, are employed to generate different types of explanations. The model achieved an accuracy of 82.54% on the initial dataset and 87.5% on the field dataset. XAI evaluation results indicated an accuracy of 57.5% for the LIME method, 55% for the SHAP method, and 67.5% for the Grad-CAM method. Based on this evaluation, the LIME method was identified as the most effective for providing model explanations. The developed mobile application is capable of classifying diseases and pest symptoms on rice plant leaves, as well as explaining the classification results, with server support to facilitate this process.