Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal RABIT: Jurnal Teknologi dan Sistem Informasi Univrab
Daiyan Akbar Setiyadi
Universitas Dian Nuswantoro
Computer Science & IT Engineering

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
ANALISIS PERBANDINGAN KINERJA ARSITEKTUR RESNET50 DAN EFFICIENTNETB1 MENGGUNAKAN METODE FINE-TUNING UNTUK KLASIFIKASI PNEUMONIA Daiyan Akbar Setiyadi; Sindhu Rakasiwi
Rabit : Jurnal Teknologi dan Sistem Informasi Univrab Vol 11 No 1 (2026): Januari
Publisher : LPPM Universitas Abdurrab

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36341/rabit.v11i1.7273

Abstract

Early detection of pneumonia through chest X-ray images is a crucial step in medical treatment but is often hampered by class imbalance issues in datasets, leading to biased deep learning models. This research aims to conduct a holistic performance evaluation of two modern Convolutional Neural Network (CNN) architectures, ResNet50 and EfficientNetB1, to determine the optimal model under imbalanced data conditions. The methodology employed is transfer learning with an optimized two-phase fine-tuning protocol, supported by data augmentation techniques and class_weight strategies to address data imbalance. Evaluation was performed on the public "Chest X-Ray Images (Pneumonia)" dataset using accuracy, precision, recall, F1-score, and confusion matrix analysis. The results indicate that although ResNet50 achieved the highest total accuracy (89%) with low False Negatives (21 cases), the EfficientNetB1 model (87% accuracy) proved to be fundamentally more balanced. This superiority is demonstrated by a significant increase in the recall of the minority class (NORMAL) to 0.84, along with a 24% reduction in False Positive errors. This study concludes that a clinical trade-off exists where architecture selection must align with specific needs: ResNet50 for high-sensitivity screening, or EfficientNetB1 for prediction reliability and balance.
Co-Authors Sindhu Rakasiwi