<![CDATA[Abstrak Penelitian ini bertujuan mengembangkan model klasifikasi aksara Sunda dari citra tulisan tangan menggunakan arsitektur Inception-V3. Dataset diperoleh dari Kaggle dan melalui tahap preprocessing sebelum pelatihan model. Hasil eksperimen menunjukkan bahwa model mencapai nilai terbaik sebesar 88,3% pada konfigurasi batch-size 16 dan epoch 30. Namun, performa menurun signifikan pada data riil dengan akurasi 68,7%, yang mengindikasikan keterbatasan generalisasi model terhadap data riil. Penurunan ini disebabkan oleh kemiripan visual antar karakter, keterbatasan jumlah data, serta variasi kualitas citra. Kontribusi utama penelitian ini adalah pengembangan baseline model klasifikasi aksara Sunda berbasis deep learning menggunakan arsitektur Inception-V3, analisis komprehensif terhadap gap performa antara data dataset publik dan data riil, serta identifikasi faktor-faktor kritis yang mempengaruhi akurasi klasifikasi, seperti kemiripan visual karakter dan kualitas dataset. Kata kunci: Aksara sunda, Inception-V3, klasifikasi Abstract This study aims to develop a classification model for Sundanese script from handwritten images using the Inception-V3 architecture. The dataset was obtained from Kaggle and underwent preprocessing prior to model training. Experimental results show that the model achieved the best accuracy of 88.3% with a batch size of 16 and 30 epochs. However, performance decreased significantly on primary (real-world) data, with an accuracy of 68.7%, indicating limited model generalization. This decline is attributed to visual similarity among characters, limited data availability, and variations in image quality. The main contributions of this study include the development of a baseline deep learning model for Sundanese script classification using Inception-V3, a comprehensive analysis of the performance gap between public datasets and real-world data, and the identification of critical factors affecting classification accuracy, such as character similarity and dataset quality. Keywords: Aksara sundanese, classification, Inception-V3]]>
Copyrights © 2026
