<![CDATA[Face shape classification is an important component in various face recognition–based applications, such as security systems, social media personalization, and style recommendation systems. However, accurate face shape classification remains a challenge due to the complex variations in facial morphological structures, such as oval, square, heart, oblong, and round shapes. This study uses a Convolutional Neural Network (CNN) algorithm with the DenseNet-201 architecture. The collected dataset consists of 5,000 facial images categorized into Oval, Round, Square, Heart, and Oblong, with a data split of 80% for training and 20% for testing. The model was then trained using the DenseNet-201 architecture and evaluated using two different learning rates (0.001 and 0.0001), two batch sizes (16 and 32), and 50 and 100 epochs. The initial results showed a model accuracy of 42%, with a tendency to classify face images as the square category. Subsequently, an additional experiment was conducted, which successfully achieved an accuracy of 89%.Keywords: Facial shape; Convolutional Neural Network; DenseNet-201; Facial classification AbstrakKlasifikasi bentuk wajah merupakan komponen penting dalam berbagai aplikasi berbasis pengenalan wajah, seperti sistem keamanan, personalisasi media sosial, hingga sistem rekomendasi gaya. Namun, pengklasifikasian bentuk wajah secara akurat masih menjadi tantangan karena adanya variasi kompleks dalam struktur morfologi wajah, seperti bentuk oval, persegi, hati, oblong dan bulat. Penelitian ini menggunakan algoritma CNN dengan arsitektur DenseNet-201. Dataset yang dikumpulkan sebanyak 5000 gambar wajah, dengan kategori (Oval, Round (Bulat), Square (Persegi), Heart (Hati), dan Oblong/lonjong), dengan masing-masing proporsi 80% data Training dan 20% data Testing. Kemudian model dilatih dengan arsitektur DenseNet-201 dan dilakukan uji coba dengan dua nilai learning rate yang berbeda 0.001 dan 0.0001, dua ukuran batch 16 dan 32, dan epoch 50 dan 100. Hasil akurasi model yang di dapatkan sebesar 42% dengan kecenderungan model klasifikasi citra bentuk wajah square. Lalu dilakukan eksperimen tambahan yang berhasil mendapatkan akurasi sebesar 89%. ]]>
