Perbandingan Metode Inception dan Xception dalam Mendeteksi Penyakit Tanaman Daun Pisang

Abstract

Banana is one of Indonesia's leading agricultural commodities, known for its high nutritional value and stable productivity. However, diseases affecting banana leaves, such as Cordana, Sigatoka, and Pestalotiopsis, can significantly reduce both quality and yield. This study aims to compare the performance of two deep learning architectures, Inception and Xception, in detecting banana leaf diseases based on image data. The research methodology follows the CRISP-DM approach, including business understanding, data preparation, modeling, and evaluation stages. The dataset was obtained from Kaggle and categorized into four classes: healthy, Cordana, Sigatoka, and Pestalotiopsis. Model evaluation was carried out using accuracy, precision, recall, and f1-score metrics. The results showed that the Xception model with an undersampling scheme achieved the best performance, with an accuracy of 95.19%, precision of 83.65%, recall of 87.99%, and f1-score of 88.26%, outperforming the Inception model in all evaluation parameters. Thus, the Xception architecture is proven to be more effective in identifying banana leaf diseases based on images and has the potential to support more optimal agricultural productivity.