Penelitian ini bertujuan untuk mengimplementasikan dan membandingkan kinerja dua arsitektur Convolutional Neural Network (CNN), yaitu VGG-16 dan DenseNet-169, dalam mengklasifikasikan citra rontgen dada ke dalam dua kelas, yaitu normal dan pneumonia. Dataset yang digunakan berasal dari Kaggle, berisi 5.863 citra yang dibagi menjadi 70% data pelatihan, 20% data validasi, dan 10% data pengujian. Proses preprocessing meliputi pengubahan ukuran citra menjadi 224×224 piksel, penskalaan nilai piksel, serta penerapan berbagai teknik augmentasi pada data latih menggunakan ImageDataGenerator. Model dilatih dengan pendekatan transfer learning, di mana lapisan awal pada base model dibekukan dan lapisan klasifikasi baru ditambahkan sesuai kebutuhan klasifikasi biner.jumlah pengulangan pelatihan ditetapkan sebanyak 20, karena pada titik tersebut akurasi validasi telah optimal dan belum terjadi overfitting. Hasil evaluasi menunjukkan bahwa model VGG-16 mencapai akurasi tertinggi sebesar 90,71%, sedangkan DenseNet-169 memperoleh akurasi sebesar 90,22%. Analisis lebih lanjut terhadap metrik evaluasi seperti precision, recall, dan F1-score menunjukkan bahwa VGG-16 memiliki kinerja yang sedikit lebih unggul dibandingkan DenseNet-169 dalam penelitian ini.

This study aims to implement and compare the performance of two Convolutional Neural Network (CNN) architectures, namely VGG-16 and DenseNet-169, in classifying chest X-ray images into two categories: normal and pneumonia. The dataset used was obtained from Kaggle, consisting of 5,863 images divided into 70% training data, 20% validation data, and 10% testing data. The preprocessing stage included resizing images to 224×224 pixels, scaling pixel values, and applying various augmentation techniques to the training data using ImageDataGenerator. The models were trained using a transfer learning approach, in which the initial layers of the base models were frozen, and new classification layers were added to suit binary classification tasks. The training was conducted for 20 epochs, as the validation accuracy had reached an optimal level without signs of overfitting. The evaluation results show that the VGG-16 model achieved the highest accuracy of 90.71%, while the DenseNet-169 model obtained 90.22%. Further analysis of evaluation metrics such as precision, recall, and F1-score revealed that VGG-16 performed slightly better than DenseNet-169 in this study.