Skripsi
MODEL ENSEMBLE MODEL DEEP LEARNING UNTUK MENINGKATKAN PERFORMA KLASIFIKASI PNEUMONIA PADA CITRA X-RAY DADA
Pneumonia is a lung infection that requires fast and accurate diagnosis. This study aims to improve pneumonia classification performance on Chest X-ray (CXR) images using ensemble learning methods. Three Convolutional Neural Network (CNN) models, namely DenseNet121, MobileNetV2, and Xception, were employed as base models using a transfer learning approach. The ensemble methods applied include hard voting, soft voting, and stacking with logistic regression as the meta-learner. The experimental results show that DenseNet121, MobileNetV2, and Xception achieved accuracies of 95.56%, 95.22%, and 95.11%, respectively. Hard voting and soft voting achieved an accuracy of 96.13%, while stacking delivered the best performance across all evaluation metrics, achieving an accuracy of 97.04% along with higher precision, recall, and F1-score compared to the other methods. The results indicate that ensemble methods, particularly stacking, can improve pneumonia classification performance compared to single models.