Manuscript received September 13, 2024; revised October 12, 2024; accepted December 2, 2024; published June 12, 2025.

Abstract—In the era of health digitalization, early detection of pneumonia through medical image analysis is one of the main challenges in improving the quality of health services. This study aims to enhance pneumonia classification performance using Convolutional Neural Network (CNN) architecture optimization and careful hyperparameter tuning. Through the application of optimization techniques such as Random Search, Bayesian Optimization, and tuning key hyperparameters such as the number of convolution layers, kernel size, dropout rate, and learning rate, this research succeeded in identifying the optimal model configuration. The Proposed Method model shows the best overall performance based on research results involving three models: Proposed Method, VGG16, and ResNet50. With the highest F1-Score value of 0.8440, accuracy of 0.9000, and lowest loss of 0.0977, the Proposed Method achieved an optimal balance between recall and Precision. Although VGG16 has the highest recall, its low precision value shows a tendency to produce more false positives. In contrast, the Proposed Method, with the best Precision of 0.7600 and superior accuracy performance, makes it the most reliable model for classifying pneumonia in this study. Experimental results show a significant increase in classification accuracy compared with conventional approaches, thus supporting further implementation in clinical applications. This study also provides insight into the importance of a systematic approach in designing and optimizing CNN models for disease classification tasks, especially pneumonia.

Keywords—pneumonia classification, Convolutional Neural Network (CNN), hyperparameter tuning, CNN architecture optimization, medical imaging, deep learning, image classification

Cite: Solikhun, Mochamad Wahyudi, and Agus Perdana Windarto, "Enhancing Pneumonia Classification Performance through CNN Architecture Optimization and Hyperparameter Tuning," Journal of Image and Graphics, Vol. 13, No. 3, pp. 286-292, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.